https://wiki.mobileread.com/wiki/MOBI#Format

Like PalmDOC, the Mobipocket file format is that of a standard Palm Database Format file. The header of that format includes the name of the database (usually the book title and sometimes a portion of the authors name) which is up to 31 bytes of data. The files are identified as Creator ID of MOBI and a Type of BOOK.

Mobipocket have some minimal file format info, mainly about the HTML encoding they use in the text of the book, at http://www.mobipocket.com/dev/article.asp?BaseFolder=prcgen

[edit]PalmDOC Header

The first record in the Palm Database Format gives more information about the Mobipocket file. The first 16 bytes are almost identical to the first sixteen bytes of a PalmDOC format file.

offset	bytes	content	comments
0	2	Compression	1 == no compression, 2 = PalmDOC compression, 17480 = HUFF/CDIC compression
2	2	Unused	Always zero
4	4	text length	Uncompressed length of the entire text of the book
8	2	record count	Number of PDB records used for the text of the book.
10	2	record size	Maximum size of each record containing text, always 4096
12	4	Current Position	Current reading position, as an offset into the uncompressed text

There are two differences from a Palm DOC file. There’s an additional compression type (17480), and the Current Position bytes are used for a different purpose:

offset	bytes	content	comments
12	2	Encryption Type	0 == no encryption, 1 = Old Mobipocket Encryption, 2 = Mobipocket Encryption
14	2	Unknown	Usually zero

The old Mobipocket Encryption scheme only allows the file to be registered with one PID, unlike the current encryption scheme that allows multiple PIDs to be used in a single file. Unless specifically mentioned, all the encryption information on this page refers to the current scheme.

[edit]MOBI Header

Most Mobipocket file also have a MOBI header in record 0 that follows these 16 bytes, and newer formats also have an EXTH header following the MOBI header, again all in record 0 of the PDB file format.

The MOBI header is of variable length and is not documented. Some fields have been tentatively identified as follows:

offset	hex	bytes	content	comments
16	0x10	4	identifier	the characters M O B I
20	0x14	4	header length	the length of the MOBI header, including the previous 4 bytes
24	0x18	4	Mobi type	The kind of Mobipocket file this is 2 Mobipocket Book 3 PalmDoc Book 4 Audio 232 mobipocket? generated by kindlegen1.2 248 KF8: generated by kindlegen2 257 News 258 News_Feed 259 News_Magazine 513 PICS 514 WORD 515 XLS 516 PPT 517 TEXT 518 HTML
28	0x1c	4	text Encoding	1252 = CP1252 (WinLatin1); 65001 = UTF-8
32	0x20	4	Unique-ID	Some kind of unique ID number (random?)
36	0x24	4	File version	Version of the Mobipocket format used in this file.
40	0x28	4	Ortographic index	Section number of orthographic meta index. 0xFFFFFFFF if index is not available.
44	0x2c	4	Inflection index	Section number of inflection meta index. 0xFFFFFFFF if index is not available.
48	0x30	4	Index names	0xFFFFFFFF if index is not available.
52	0x34	4	Index keys	0xFFFFFFFF if index is not available.
56	0x38	4	Extra index 0	Section number of extra 0 meta index. 0xFFFFFFFF if index is not available.
60	0x3c	4	Extra index 1	Section number of extra 1 meta index. 0xFFFFFFFF if index is not available.
64	0x40	4	Extra index 2	Section number of extra 2 meta index. 0xFFFFFFFF if index is not available.
68	0x44	4	Extra index 3	Section number of extra 3 meta index. 0xFFFFFFFF if index is not available.
72	0x48	4	Extra index 4	Section number of extra 4 meta index. 0xFFFFFFFF if index is not available.
76	0x4c	4	Extra index 5	Section number of extra 5 meta index. 0xFFFFFFFF if index is not available.
80	0x50	4	First Non-book index?	First record number (starting with 0) that’s not the book’s text
84	0x54	4	Full Name Offset	Offset in record 0 (not from start of file) of the full name of the book
88	0x58	4	Full Name Length	Length in bytes of the full name of the book
92	0x5c	4	Locale	Book locale code. Low byte is main language 09= English, next byte is dialect, 08 = British, 04 = US. Thus US English is 1033, UK English is 2057.
96	0x60	4	Input Language	Input language for a dictionary
100	0x64	4	Output Language	Output language for a dictionary
104	0x68	4	Min version	Minimum mobipocket version support needed to read this file.
108	0x6c	4	First Image index	First record number (starting with 0) that contains an image. Image records should be sequential.
112	0x70	4	Huffman Record Offset	The record number of the first huffman compression record.
116	0x74	4	Huffman Record Count	The number of huffman compression records.
120	0x78	4	Huffman Table Offset
124	0x7c	4	Huffman Table Length
128	0x80	4	EXTH flags	bitfield. if bit 6 (0x40) is set, then there’s an EXTH record
132	0x84	32	?	32 unknown bytes, if MOBI is long enough
164	0xa4	4	Unknown	Use 0xFFFFFFFF
168	0xa8	4	DRM Offset	Offset to DRM key info in DRMed files. 0xFFFFFFFF if no DRM
172	0xac	4	DRM Count	Number of entries in DRM info. 0xFFFFFFFF if no DRM
176	0xb0	4	DRM Size	Number of bytes in DRM info.
180	0xb4	4	DRM Flags	Some flags concerning the DRM info.
184	0xb8	8	Unknown	Bytes to the end of the MOBI header, including the following if the header length >= 228 (244 from start of record). Use 0x0000000000000000.
192	0xc0	2	First content record number	Number of first text record. Normally 1.
194	0xc2	2	Last content record number	Number of last image record or number of last text record if it contains no images. Includes Image, DATP, HUFF, DRM.
196	0xc4	4	Unknown	Use 0x00000001.
200	0xc8	4	FCIS record number
204	0xcc	4	Unknown (FCIS record count?)	Use 0x00000001.
208	0xd0	4	FLIS record number
212	0xd4	4	Unknown (FLIS record count?)	Use 0x00000001.
216	0xd8	8	Unknown	Use 0x0000000000000000.
224	0xe0	4	Unknown	Use 0xFFFFFFFF.
228	0xe4	4	First Compilation data section count	Use 0x00000000.
232	0xe8	4	Number of Compilation data sections	Use 0xFFFFFFFF.
236	0xec	4	Unknown	Use 0xFFFFFFFF.
240	0xf0	4	Extra Record Data Flags	A set of binary flags, some of which indicate extra data at the end of each text block. This only seems to be valid for Mobipocket format version 5 and 6 (and higher?), when the header length is 228 (0xE4) or 232 (0xE8). bit 1 (0x1): <extra multibyte bytes><size> bit 2 (0x2): <TBS indexing description of this HTML record><size> bit 3 (0x4): <uncrossable breaks><size> Setting bit 2 (0x2) disables <guide><reference type=”start”> functionality.
244	0xf4	4	INDX Record Offset	(If not 0xFFFFFFFF)The record number of the first INDX record created from an ncx file.
248	0xf8	4	Unknown	0xFFFFFFFF In new MOBI file, the MOBI header length is 256, skip this to EXTH header.
252	0xfb	4	Unknown	0xFFFFFFFF In new MOBI file, the MOBI header length is 256, skip this to EXTH header.
256	0x100	4	Unknown	0xFFFFFFFF In new MOBI file, the MOBI header length is 256, skip this to EXTH header.
260	0x104	4	Unknown	0xFFFFFFFF In new MOBI file, the MOBI header length is 256, skip this to EXTH header.
264	0x108	4	Unknown	0xFFFFFFFF In new MOBI file, the MOBI header length is 256, skip this to EXTH header.
268	0x10b	4	Unknown	0 In new MOBI file, the MOBI header length is 256, skip this to EXTH header, MOBI Header length 256, and add 12 bytes from PalmDOC Header so this index is 268.

[edit]EXTH Header

If the MOBI header indicates that there’s an EXTH header, it follows immediately after the MOBI header. Since the MOBI header is of variable length, this isn’t at any fixed offset in record 0. Note that some readers will ignore any EXTH header info if the mobipocket version number specified in the MOBI header is 2 or less (perhaps 3 or less).

The EXTH header is also undocumented, so some of this is guesswork.

bytes	content	comments
4	identifier	the characters E X T H
4	header length	the length of the EXTH header, including the previous 4 bytes – but not including the final padding.
4	record Count	The number of records in the EXTH header. the rest of the EXTH header consists of repeated EXTH records to the end of the EXTH length.
	EXTH record start	Repeat until done.
4	record type	Exth Record type. Just a number identifying what’s stored in the record
4	record length	length of EXTH record = L , including the 8 bytes in the type and length fields
L-8	record data	Data.
	EXTH record end	Repeat until done.
p	padding	Null bytes to pad the EXTH header to a multiple of four bytes (none if the header is already a multiple of four). This padding is not included in the EXTH header length.

There are lots of different EXTH Records types. Ones found so far in Mobipocket files are listed here, with possible meanings. Hopefully the table will be filled in as more information comes to light.

record type	usual length	name	comments	opf meta tag
1		drm_server_id
2		drm_commerce_id
3		drm_ebookbase_book_id
100		author		<dc:Creator>
101		publisher		<dc:Publisher>
102		imprint		<Imprint>
103		description		<dc:Description>
104		isbn		<dc:Identifier scheme=’ISBN’>
105		subject	Could appear multiple times	<dc:Subject>
106		publishingdate		<dc:Date>
107		review		<Review>
108		contributor		<dc:Contributor>
109		rights		<dc:Rights>
110		subjectcode		<dc:Subject BASICCode=”subjectcode”>
111		type		<dc:Type>
112		source		<dc:Source>
113		asin	Kindle Paperwhite labels books with “Personal” if they don’t have this record.
114		versionnumber
115	4	sample	0x0001 if the book content is only a sample of the full book
116		startreading	Position (4-byte offset) in file at which to open when first opened
117	3	adult	Mobipocket Creator adds this if Adult only is checked on its GUI; contents: “yes”	<Adult>
118		retail price	As text, e.g. “4.99”	<SRP>
119		retail price currency	As text, e.g. “USD”	<SRP Currency=”currency”>
121	4	KF8 BOUNDARY Offset
125	4	count of resources
129		KF8 cover URI
131	4	Unknown
200	3	Dictionary short name	As text	<DictionaryVeryShortName>
201	4	coveroffset	Add to first image field in Mobi Header to find PDB record containing the cover image	<EmbeddedCover>
202	4	thumboffset	Add to first image field in Mobi Header to find PDB record containing the thumbnail cover image
203		hasfakecover
204	4	Creator Software	Known Values: 1=mobigen, 2=Mobipocket Creator, 200=kindlegen (Windows), 201=kindlegen (Linux), 202=kindlegen (Mac). Warning: Calibre creates fake creator entries, pretending to be a Linux kindlegen 1.2 (201, 1, 2, 33307) for normal ebooks and a non-public Linux kindlegen 2.0 (201, 2, 0, 101) for periodicals.
205	4	Creator Major Version
206	4	Creator Minor Version
207	4	Creator Build Number
208		watermark
209		tamper proof keys	Used by the Kindle (and Android app) for generating book-specific PIDs.
300		fontsignature
401	1	clippinglimit	Integer percentage of the text allowed to be clipped. Usually 10.
402		publisherlimit
403		Unknown
404	1	ttsflag	1 – Text to Speech disabled; 0 – Text to Speech enabled
405	1	Unknown (Rent/Borrow flag?)	1 in this field seems to indicate a rental book
406	8	Rent/Borrow Expiration Date	If this field is removed from a rental, the book says it expired in 1969
407	8	Unknown
450	4	Unknown
451	4	Unknown
452	4	Unknown
453	4	Unknown
501	4	cdetype	PDOC – Personal Doc; EBOK – ebook; EBSP – ebook sample;
502		lastupdatetime
503		updatedtitle
504		asin	I found a copy of ASIN in this record.
524		language		<dc:language>
525		alignment	I found horizontal-lr in this record.
535		Creator Build Number	I found 1019-d6e4792 in this record, which is a build number of Kindlegen 2.7
547		InMemory	String ‘Ix00nx00Mx00ex00mx00ox00rx00yx00’ found in this record, for KindleGen V2.9 build 1029-0897292

[edit]Remainder of Record 0

At the end of Record 0 of the PDB file format, we usually get the full file name, the offset of which is given in the MOBI header.

There might be data of unknown use between the end of the EXTH records and the name.

The name is followed by two null bytes, and then padded with null bytes to a four-byte boundary. For example, if the name is 16 bytes long, with two null bytes, that makes 18 bytes, and it then gets another two null bytes added to make it up to 20 bytes in total. However, the length stored in the header is only 16. If the name was 19 bytes, it would be followed by two null bytes to make it up to 21 bytes, and then padded with three more null bytes to make it up to 24 bytes.

The name and padding is followed by more data of unknown use, usually null bytes, to the end of section 0.

[edit]Index meta record

The first record of an index contains the meta data of the index.

offset	hex	bytes	content	comments
0	0x00	4	Identifier	the characters I N D X
4	0x04	4	header length	the length of the INDX header, including the previous 4 bytes
8	0x08	4	index type	the type of the index. Known values: 0 – normal index, 2 – inflections
12	0x0c	4	?	?
16	0x10	4	?	?
20	0x14	4	idxt start	the offset to the IDXT section
24	0x18	4	index count	the number of index records
28	0x1c	4	index encoding	1252 = CP1252 (WinLatin1); 65001 = UTF-8
32	0x20	4	index language	the language code of the index
36	0x24	4	total index count	the number of index entries
40	0x28	4	ordt start	the offset to the ORDT section
44	0x2c	4	ligt start	the offset to the LIGT section
48	0x30	4	?	?
52	0x34	4	?	?

The remaining INDX header values are unknown.

[edit]TAGX section

The TAGX section follows the INDX header and is essential for decoding the index values, as it defines which how many control bytes an entry contains, which bits correspond to which tag and how many values a tag requires (most tag need one value, but some have two, maybe more).

offset	hex	bytes	content	comments
0	0x00	4	Identifier	the characters T A G X
4	0x04	4	header length	the length of the TAGX header, including the previous 4 bytes
8	0x08	4	control byte count	the number of control bytes
12	0x0c	n	tag table	the tag table entries (n = header length – 12, must be multiple of 4 bytes)

The tag table entries are multiple of 4 bytes. The first byte is the tag, the second byte the number of values, the third byte the bit mask and the fourth byte indicates the end of the control byte. If the fourth byte is 0x01, all other bytes of the entry are zero.

[edit]Variable-width integers

Some parts of the Mobipocket format encode data as variable-width integers. These integers are represented big-endian with 7 bits per byte in bits 1-7. They may be either forward-encoded, in which case only the LSB has bit 8 set, or backward-encoded, in which case only the MSB has bit 8 set. For example, the number 0x11111 would be represented forward-encoded as:

   0x04 0x22 0x91

And backward-encoded as:

   0x84 0x22 0x11

[edit]Trailing entries

The Extra Data Flags field of the MOBI header indicates which, if any, trailing entries are appended to the end of each text record. Each set bit in the field indicates a trailing entry. The entries appear to occur in bit-order; e.g., trailing entry 1 immediately follows the text content and entry 16 occurs at the very end of the record. The effect and exact details of most of these entries is unknown. The trailing entries indicated by bits 2-16 appear to follow a common format. That format is:

   <data><size>

Where <size> is the size of the entire trailing entry (including the size of <size>) as a backward-encoded Mobipocket variable-width integer.

Only a few bits have been identified

bit	Data at end of records
0x0001	Multi-byte character overlaps
0x0002	Some data to help with indexing
0x0004	Some data about uncrossable breaks

[edit]Multibyte character overlap

When bit 1 of the Extra Data Flags field is set, each record is followed by a trailing entry containing any extra bytes necessary to complete a multibyte character which crosses the record boundary. The bytes do not participate in compression regardless which compression scheme is used for the file. However, unlike the trailing data bytes, the multibytes (including the count byte) do get included in any encryption. The overlapping bytes then re-appear as normal content at the beginning of the following record. The trailing entry ends with a byte containing a count of the overlapping bytes plus additional flags.

offset	bytes	content	comments
0	0-3	N terminal bytes of a multibyte character
N	1	Size & flags	bits 1-2 encode N, use of bits 3-8 is unknown

[edit]PalmDOC Compression

PalmDOC uses LZ77 compression techniques, an implementation for PalmDOC can be found at Github . DOC files can contain only compressed text. The format does not allow for any text formatting. This keeps files small, in keeping with the Palm philosophy. However, extensions to the format can use tags, such as HTML or PML, to include formatting within text. These extensions to PalmDoc are not interchangeable and are the basis for most eBook Reader formats on Palm devices.

LZ77 algorithms achieve compression by replacing portions of the data with references to matching data that has already passed through both encoder and decoder. A match is encoded by a pair of numbers called a length-distance pair, which is equivalent to the statement “each of the next length characters is equal to the character exactly distance characters behind it in the uncompressed stream.” (The “distance” is sometimes called the “offset” instead.)

In the PalmDoc format, a length-distance pair is always encoded by a two-byte sequence. Of the 16 bits that make up these two bytes, 11 bits go to encoding the distance, 3 go to encoding the length, and the remaining two are used to make sure the decoder can identify the first byte as the beginning of such a two-byte sequence. The exact algorithm needed to decode the compressed text can be found on the PalmDOC page.

PalmDOC data is always divided into 4096 byte blocks (uncompressed size) and the blocks are acted upon independently; no information from previous or later blocks is needed when a block is being compressed or decompressed.

PalmDOC does have support for bookmarks. These pointers are named and refer to an offset location in a file. If the file is edited these locations may no longer refer to the correct locations. Some reading programs allow the user to enter or edit these bookmarks while others treat them as a TOC. Some reading programs may ignore them entirely. They are stored at the end of the file itself so the full file needs to be scanned when loaded to find them.

[edit]Image Records

If the file contains images, they follow the text blocks, with each image using a single block. The 4096-byte record size in the PalmDoc header applies only to text records; image records may be larger.

[edit]Magic Records

In some cases, MobiPocket Creator adds a 2-zero-byte record after the text records in a file. This record is not included in the “record count” of text records in the PalmDoc header, and is also not used as the “first non-book index” in the MOBI header. (If the 2-zero-byte record is present, the index of the following block is used as the “first non-book index”.)

MobiPocket Creator also ends files with three records: ‘FLIS’, ‘FCIS’, and ‘end-of-file’, in that order. The ‘FLIS’ and ‘FCIS’ records do not seem to be necessary for MobiPocket Reader or the Amazon Kindle 2to read the file. The ‘end-of-file’ record might be necessary.

[edit]FLIS Record

The FLIS record appears to have a fixed value. The meaning of the values is not known.

offset	bytes	content	comments
0	4	identifier	the characters F L I S (0x46 0x4c 0x49 0x53)
4	4	?	fixed value: 8
8	2	?	fixed value: 65
10	2	?	fixed value: 0
12	4	?	fixed value: 0
16	4	?	fixed value: -1 (0xFFFFFFFF)
20	2	?	fixed value: 1
22	2	?	fixed value: 3
24	4	?	fixed value: 3
28	4	?	fixed value: 1
32	4	?	fixed value: -1 (0xFFFFFFFF)

[edit]FCIS Record

The FCIS record appears to have mostly fixed values.

offset	bytes	content	comments
0	4	identifier	the characters F C I S (0x46 0x43 0x49 0x53)
4	4	?	fixed value: 20
8	4	?	fixed value: 16
12	4	?	fixed value: 1
16	4	?	fixed value: 0
20	4	?	text length (the same value as “text length” in the PalmDoc header)
24	4	?	fixed value: 0
28	4	?	fixed value: 32
32	4	?	fixed value: 8
36	2	?	fixed value: 1
38	2	?	fixed value: 1
40	4	?	fixed value: 0

[edit]End-of-file Record

The end-of-file record is a fixed 4-byte record. While the last two bytes appear to be a CRLF marker, the meaning of the first two bytes is unknown.

offset	bytes	content	comments
0	1	?	fixed value: 233 (0xe9)
1	1	?	fixed value: 142 (0x8e)
2	1	?	fixed value: 13 (0x0d)
3	1	?	fixed value: 10 (0x0a)

[edit]Compilation Records

KindleGen creates records of the compilation source (KindleGen 1.2-2.5) and the compilation source and compiler output (Kindle Gen 2.7-) just before the #End-of-file Record (KindleGen version 1.2-2.2), or just before the BOUNDARY record (KindleGen version 2.3-).

MOBI files created with Mobipocket creator, Amazon’s Personal Document Service, or Kindle Direct Publishing (former Amazon DTP) don’t include SRCS record. In a past, kindlegen had an undocumented option to suppress this record, but the option was removed in 2010.

A SRCS record is a record whose content is a zip archive of all source files (i.e., .opf, .ncx, .htm, .jpg, …) given to the command and puts it in the generated MOBI file. The record begins with the “SRCS” signature and looks as follows:

offset	bytes	content	comments
0	4	identifier	“SRCS” (0x53 0x52 0x43 0x53)
4	4	?	fixed value(?): 0x00000010
8	4	?	fixed value(?): 0x0000002f
12	4	?	fixed value(?): 0x00000001
16		zip	The zip archive continues to the end of this record

A CMET record is a record whose content is the output of the compilation operation, and perhaps extra info. The record begins with the “CMET” signature and looks as follows:

offset	bytes	content	comments
0	4	identifier	“CMET” (0x43 0x4D 0x45 0x54)
4	4	?	fixed value(?): 0x0000000C
8	4	text length	(big endian)
12	variable	text	compilation output text, line endings are CRLF
variable	variable	?	unknown data to the end of the record

[edit]Media Records (AUDI/VIDE)

kindlegen supports embedded audio and video for some Kindle platforms. Each media file is stored in a separate AUDI (audio) or VIDE (video) record.

A media record looks as follows:

offset	bytes	content	comments
0	4	identifier	“AUDI” (0x41 0x55 0x44 0x49) or “VIDE” (0x56 0x49 0x44 0x45)
4	4	?	unkown value
8	4	?	unknown value
12		media	The media data continues to the end of this record

[edit]MBP

This is the extension used on a side file (auxiliary) for MOBI formatted eBooks. It is used to store metadata used by the library software and also to store user entered data like bookmarks, annotations, last read position. This file is created automatically by the reader program when the eBook is first opened and has a .mbp extension. The Library management software in MobiPocket uses this file to get information displayed in the library window such as title, author, and description so that it won’t have to open the larger eBook file.

There is an ongoing effort to describe the binary MBP file format (see this site). There is also a mbp reader program that will extract notes from a mbp file

[edit]eBook Creation

There are several ways to create eBooks in the MOBI format. The rules for the format of the source files need to create eBooks in MOBI are spelled out in documents on the MobiPocket web site. The recommended tool called MobiPocket Creator is available as a download from the web site.

EBooks can also be converted from other forms using the Windows version of the MobiPocket Reader. Once converted the file can be used on any device supported by MobiPocket Reader.

[edit]Guidelines

In order to better support the features of the MobiPocket Reader there are some guidelines that need to be followed when creating a book in this format.

Do not specify a default font family, font size or other font attributes such as weight or color. This is a choice the person reading the eBook should be able to make. Fonts Sizes and Attributes can be specified for special headings and other specific items. Use only generic font families.
Do not impose justification for standard text. It may be needed for captions and other special text.
Do not use tables for anything except table data. Nested tables are not supported.
Do not use blank lines to try and force page changes. Use the <mbp:pagebreak/> tag.
Do not use multiple books for different devices. Instead use advanced features such as multi resolution images and platform specific frames.

[edit]Adapting images to various PDA screen resolutions

Note that the following section only applies to the original mobi format and is not used by Amazon AZW files.

The IMG tag in Mobipocket publications supports up to three source attributes for various resolutions: src, losrc and hisrc. This makes it possible to optimize the same ebook for various devices. The image to be displayed is dynamically selected by the Reader according to the resolution of the screen on the actual device:

Attribute	screen smallest size	example devices
losrc	<= 239 pixels	Low rez 160×160 Palm devices (PalmVx, Treo 600, Zire) Smartphones (Nokia 3650, Sony Ericson P800/900, Microsoft smartphones)
src	>= 240 pixels (handhelds)	Pocket PC, Hi rez Palm devices (Sony Clie, Tungsten, Zire 71)
hisrc	>= 480 pixels	any desktop or tablet PC

Example:

<img hisrc="cover480x640.gif" src="cover220x300.gif" losrc="cover140x140.gif"/>

Please also notice that there is a 63KB internal limitation for images (this is a restriction of the Mobipocket .PRC format). GIFs have to be smaller than 63KB. You can use GIF optimization programs such as Ulead Smart Saver to get GIFs smaller than 63KB. (If images are bigger than 63KB, they are automatically resized to fit in the limit by MobiGEN but you might not like the result). Jpeg images will use a lower Quality setting to get the image size down without reducing the pixel size.

[edit]HTML and CSS Tips for MOBI creation

Kindle and Kindle DX do not handle the soft hyphen HTML entity correctly. Use the <shy/> tag instead.
Grey text is displayed as white on some devices. To avoid this problem, add:

@media amzn-mobi and (monochrome) {.mygreytextclass {color: black;}}

The kindlegen tool ignores padding-left. If necessary, you can work around this by adding an element inside that element and setting its left margin.
The kindlegen tool’s CSS parser is sometimes buggy. As a result, if you have:

div.foo p {…}the kindlegen tool often incorrectly interprets it as:div.foo, div.foo p {…}In situations where you want to apply the style to only the inner tag, you must add a custom class to the paragraphs inside the outer tag and use that CSS selector by itself.

For additional tips specific to developing content for KF8-capable readers, see the KF8 CSS Tips.