Convert Parquet to CSV
Upload a .parquet file and get back a CSV file. Types are preserved rather than flattened to text. Free, no registration, up to 10MB.
Click to upload or drag and drop
CSV, Excel, SQLite, SQL, DBF (max 10 MB)
How your data is typed in CSV
Generated from the converter's own CSV type mapping — this is what it actually emits, not an example.
| Your data | CSV type |
|---|---|
| Whole number | text (CSV has no types) |
| Decimal (money) | text (CSV has no types) |
| Floating point | text (CSV has no types) |
| True / false | text (CSV has no types) |
| Date | text (CSV has no types) |
| Date and time | text (CSV has no types) |
| Short text | text (CSV has no types) |
| Long text | text (CSV has no types) |
| Binary data | text (CSV has no types) |
Working from a .parquet file
Parquet is the one source here that arrives with a complete, machine-readable schema, so the conversion is a translation rather than an investigation. The interesting question is what Parquet can express that a SQL table cannot.
- →Lists, structs and maps have no scalar equivalent in a relational target, so those columns are carried across as JSON text instead of being exploded into invented columns.
- →Integer widths are kept rather than widened: int8 and int16 become a 16-bit column and int32 a 32-bit one, so the target schema stays as tight as the file was.
- →A Parquet dataset is usually a directory of part files with partition keys encoded in directory names like year=2024/. One file is read, and a column that exists only as a directory name is not inside it.
- →Rows are streamed in batches of 10,000, so a wide file never has to be held in memory all at once.
The shape of the CSV files you get
CSV is where the type information stops, so what is left to describe is the file itself: how a row is terminated, which fields are quoted, and how a value that used to have a type is spelled out as characters.
- →Rows end with a carriage return and a line feed, the RFC 4180 ending, while a newline inside a quoted field stays the single character it always was. The encoding is UTF-8 with no byte-order mark in front of it.
- →Quotes go on only where they are needed - a field holding a comma, a double quote or a newline - and a quote inside a field is doubled rather than escaped with a backslash. Every other field is written bare.
- →Decimal values are written digit for digit and never through a floating-point number, so a money column that was exact in the source is still exact in the file. Booleans are written lower case, as true and false, which is the spelling this converter's own reader takes back as a boolean.
Reading Parquet
- →Parquet carries a full schema, so nothing is guessed — types, decimal precision and nullability are read directly.
- →Decimal columns keep their exact precision and scale.
Writing CSV
- →One file per table, with a header row.
- →NULL is written as an empty field, never the word "None".
- →Binary data is written as hex, which is reversible.
What you get
A ZIP containing <table>.csv. If anything about your data needed a judgement call, a _warnings.txt explains it.
Questions
- Is converting Parquet to CSV free?
- Yes. Files up to 10MB convert free with no account. There is a limit of five conversions per day per IP address.
- Will my Parquet data keep its types in CSV?
- Yes — that is the point. Whole numbers, decimals, dates and booleans are mapped to real CSV types rather than everything becoming text. The mapping table above shows exactly what you get.
- Do you store my file?
- No. The file is written to a temporary directory, converted, and the directory is deleted as soon as your download has been sent.