Split CSV

In rows per file the header is not counted, so 10,000 means 10,000 records plus the header line. The 1,000,000 button exists because an Excel sheet holds 1,048,576 rows; parts of a million rows always open completely in Excel.

Number of files waits for the row count to finish, then shares the rows out evenly. 1,941,631 rows into 7 files gives six files of 277,376 rows and one of 277,375.

Max file size counts real bytes, including the header and the BOM, and starts a new file before a record would push it over the limit. In a test with a 100 MB file and a 9.5 MB limit, the largest part came out at 9,961,470 bytes against a limit of 9,961,472. The only case where a part can be larger is a single record that is bigger than the limit on its own.

Splitting a CSV by counting lines in a text editor goes wrong as soon as a field contains a line break, which is common in address and comment columns. This tool follows the CSV rules instead:

• A line break inside double quotes, such as "Flat 2
  12 High Street", belongs to the field, so that record stays in one piece in one file.
• Commas inside quotes and doubled quotes (""like this"") are left exactly as they are.
• Windows (CRLF), Unix (LF) and old Mac (CR) line endings are all recognised, and the bytes are copied unchanged, so each part keeps the line endings of the original.
• Empty lines are left out and counted in the result, so you know if any were there.
• A UTF-8 BOM at the start of the input is removed from the data and, if you keep the BOM option on, written once at the start of every part.

Because the tool only looks for quotes and line breaks, it works with any delimiter (comma, semicolon, tab, pipe) and does not need to decode the text. Files saved in Korean CP949/EUC-KR or Japanese Shift_JIS are split byte for byte and stay in their original encoding.

The file is never loaded into memory in one go. It is read in 8 MB slices, each slice is scanned for record boundaries and written straight into the ZIP using the browser's built-in compression. Memory use is roughly the size of the compressed ZIP plus one slice.

In our test, a 100 MB UTF-8 CSV with 1,941,631 records, quoted commas, doubled quotes and multi-line fields was counted in about 2 seconds and split into four files in about 9 seconds in Chromium on a desktop computer. Every part was then read back with a standard CSV parser and compared record by record with the original: same header, same records, same order, nothing lost.

The upload box accepts files up to 2 GB, and the ZIP itself must stay under 4 GB. Phones and tablets have far less memory; files of a few hundred MB may fail there, so use a computer for very large files.

Excel guesses the encoding of a CSV when you double-click it, and for UTF-8 files without a byte order mark it usually guesses wrong, which turns é into é and Korean into unreadable symbols. Keeping Add UTF-8 BOM on prevents that. Leave it off only if the parts are going to a program that rejects a BOM, such as some database import scripts.

Excel also changes values as it opens a CSV: leading zeros disappear from postcodes and phone numbers, and long IDs turn into 1.23E+15. Splitting does not change any values, but if a part needs to go into Excel cleanly, run it through CSV to Excel, which keeps those values as text.

A warning says a double quote is never closed. Somewhere in the data there is a stray quote, typically a measurement such as 27" monitor in an unquoted field. Everything after it is treated as one field, so the row count will look far too low. Fix the source file, or export it again with all fields quoted.

The preview shows everything in one column. The delimiter could not be recognised from the first lines. This only affects the preview: splitting works on records, not columns, so the parts are still correct.

The tool says the file is UTF-16. Excel's Unicode Text format saves two bytes per character. Open the file in Excel or a text editor and save it as CSV UTF-8 first.

There is one part fewer than expected. The header row is not counted as data, and empty lines are skipped. A file with 1,001 lines including the header has 1,000 records, which fit in one file of 1,000 rows.

To split a spreadsheet rather than a CSV, use Split Excel, which also splits by sheet. To turn a part into a proper workbook, use CSV to Excel; to go the other way, Excel to CSV. If you only need to look inside the ZIP or pull out one part, Unzip opens it in the browser.