.CSV File Format – Grammar and dialects

Comma-separated values

RFC 4180 (October 2005) describes the format “that seems to be followed by most implementations” and registers text/csv; it is informational, not a standard every program follows. W3C’s CSV on the Web recommendations describe the variations. This page follows RFC 4180, RFC 7111, Metadata Vocabulary for Tabular Data and Model for Tabular Data and Metadata on the Web, read 2026-11-12.

RFC 4180 grammar

file        = [header CRLF] record *(CRLF record) [CRLF]
header      = name *(COMMA name)
record      = field *(COMMA field)
name        = field
field       = (escaped / non-escaped)
escaped     = DQUOTE *(TEXTDATA / COMMA / CR / LF / 2DQUOTE) DQUOTE
non-escaped = *TEXTDATA
COMMA       = %x2C
CR          = %x0D
DQUOTE      = %x22
LF          = %x0A
CRLF        = CR LF
TEXTDATA    = %x20-21 / %x23-2B / %x2D-7E
  • Records end with CR LF; the last one may or may not.
  • Every record should have the same number of fields, and the last field must not be followed by a comma.
  • A field containing a comma, a double quote or a line break should be in double quotes; inside quotes a double quote must be written twice: "say ""hi""". An unquoted field may not contain double quotes.
  • Spaces are part of a field and should not be ignored.
  • TEXTDATA allows only printable ASCII. RFC 4180 says common usage is US-ASCII, with other character sets named by the charset parameter; the W3C model notes that this restriction “is routinely ignored in practice”.

Media type

text/csv, with optional parameters charset (such as utf-8) and header (present or absent). RFC 7111 (January 2014, updates RFC 4180) adds fragment identifiers for rows, columns and cells, such as #row=5-7, #col=2 and #cell=4,1 (row 4, column 1); rows and columns count from 1.

Dialects

W3C’s “Metadata Vocabulary for Tabular Data” (Recommendation, 17 December 2015) defines dialect descriptions, hints that tell a parser how to read a CSV file, with these defaults:

PropertyDefaultMeaning
encoding"utf-8"Character encoding, one defined in the Encoding specification.
lineTerminators["\r\n", "\n"]String or array of strings that end a row.
delimiter","Field separator.
quoteChar'"'Quote character; null also sets the escape character to none.
doubleQuotetrueIf true the escape character is " (quotes doubled); if false it is a backslash.
headertruetrue sets the header row count to 1, false to 0; ignored when headerRowCount is given.
headerRowCount1Number of header rows, a non-negative integer.
skipRows0Rows to skip at the start of the file, before a header row or data.
skipColumns0Columns to skip at the start of each row.
commentPrefix"#"Prefix marking comment lines.
skipBlankRowsfalseWhether blank rows are ignored.
skipInitialSpacefalsetrue sets trimming to "start"; ignored when trim is given.
trimtrue in the property list; false in the spec’s own default dialect exampletrue, false, "true", "false", "start" or "end".

The two defaults for trim are both in section 5.9 of the recommendation: the property definition says “The default is true”, while the default dialect description printed under it has "trim": false. The “Model for Tabular Data” also gives the trim flag a default of true.

Appendix B of the “Model for Tabular Data and Metadata on the Web” records how some programs behaved when tested: Excel for Mac 2011 (English locale) saved CSV as Windows-1252 with LF line endings, quoting only cells that need it and doubling quotes, and read UTF-8 files as Windows-1252 whether or not a byte order mark was present; Google Spreadsheets downloaded CSV as UTF-8 without a byte order mark, with LF line endings. The model also has a text/tab-separated-values response set the delimiter to tab (see TSV).