Parquet to CSV Converter
Read a .parquet file without installing a data stack. The columns stored inside are written back out as plain CSV rows that a spreadsheet, a BI tool or a colleague can open straight away.
Data Converter
or drop CSV, TSV, TXT … files here ·
Max 95 MB per file· Up to 20 files· Deleted after 30 minutes
Files are processed on our servers and deleted after 30 minutes.
Character that separates columns in CSV and text files, for both reading and writing. Auto detects it from the input and writes commas.
UTF-8 works for every language. Choose UTF-8 with BOM if Excel on Windows shows garbled accents or Asian characters when opening the CSV.
Excel, ODS and Numbers input: convert the first sheet only, or every sheet (one CSV per sheet in a ZIP, or one JSON key per sheet).
Name of the sheet tab in the new spreadsheet, up to 31 characters.
Array of objects uses column names as keys for every row; array of arrays is more compact and keeps the header as the first array.
Name of the outer element that wraps all rows; each row becomes a <row> inside it.
Used in CREATE TABLE and INSERT statements. Letters, digits and underscores only.
Quoting and column types differ slightly between databases; pick the one you will import into.
Rewrite date cells in one format. ISO (2026-10-18) is safest for databases and code; keep leaves them as they appear in the source.
How to use this tool
- 1
Upload the Parquet files
Add a single .parquet file or up to 20 part files from one dataset, each up to 95 MB.
- 2
Check dates and encoding
Dates is set to ISO, so timestamps read as 2026-10-18 rather than raw numbers. Under Advanced settings, pick Delimiter Semicolon for European Excel or UTF-8 with BOM in Text encoding.
- 3
Convert to CSV
Press Convert to CSV and download. Open the file in a spreadsheet or load it wherever CSV is accepted.
Who needs Parquet as CSV
Parquet started in 2013 as a joint project of Twitter and Cloudera, built on the record-shredding ideas in Google's Dremel paper, and became an Apache top-level project in 2015. Today it is the default storage format of data lakes. BigQuery and Snowflake export to it, Spark writes it by default, and the Hugging Face Hub converts public datasets to it automatically.
That is great for engineers and awkward for everyone else. A product manager handed a .parquet file, a researcher downloading a dataset, or an accountant receiving a warehouse extract usually has no tool that opens it. CSV is the lowest common denominator that every spreadsheet and importer reads. If the people receiving the data prefer a formatted workbook, convert the CSV onward with CSV to XLSX.
Parquet vs CSV: columns on disk versus rows of text
Parquet
Binary, column by column, with the schema stored in the file footer.
- Typed values: int64, decimal, timestamp, boolean
- Nested structs and lists
- Compressed per column, read selectively
CSV
Plain text, one record per line, values separated by commas as RFC 4180 describes.
- Every value is text
- Flat: one value per cell
- Opens in any spreadsheet or editor
| from ↓ · to → | CSV | XLSX | JSON | Parquet |
|---|---|---|---|---|
| Parquet | Parquet to CSV: yes | Parquet to XLSX: yes | Parquet to JSON: yes | – Parquet to Parquet: no |
| CSV | – CSV to CSV: no | CSV to XLSX: yes | CSV to JSON: yes | CSV to Parquet: yes |
| JSON | JSON to CSV: yes | JSON to XLSX: yes | – JSON to JSON: no | JSON to Parquet: yes |
What changes when a Parquet file becomes CSV
Types are the first casualty. A decimal(18,2) amount, an int64 ID and a timestamp in UTC all become characters, and whoever opens the CSV decides how to read them again. Excel, for one, shows only 15 significant digits, so 19-digit IDs from a warehouse should be imported as text, not opened by double-click. The difference between null and an empty string disappears as well, since both are written as an empty field.
Size is the second. Columnar compression is why Parquet files are small; plain text has none, so the CSV grows. A sheet in Excel holds 1,048,576 rows, and large tables can pass that easily, so check the row count before sending the file to a spreadsheet user. Other targets, from SQL to JSON, are on the data converter page.
Apache Parquet and CSV format facts
PARQUET
Apache Parquet
Parquet is an open-source columnar storage format created by Twitter and Cloudera in 2013 and maintained by the Apache Software Foundation. It stores data by column with strong compression and typed schemas, so analytics tools can read only the columns they need. It is standard in data lakes and tools such as Spark, pandas, DuckDB, BigQuery and Snowflake. Parquet files are binary and cannot be opened in a text editor or a regular spreadsheet app.
- Compression
- Lossless
- Transparency
- –
- Animation
- –
- Introduced
- 2013 · Apache Software Foundation (from Twitter and Cloudera)
CSV
Comma-separated values
CSV is a plain-text format for tables in which each line is a row and commas separate the columns. It has been used since the early days of computing and is described in RFC 4180, though many variants use semicolons or tabs and different text encodings. Nearly every spreadsheet, database and analytics tool can import and export it, which makes it the default format for moving data between systems. CSV stores only values, with no formulas, formatting, multiple sheets or data types.
- Compression
- –
- Transparency
- –
- Animation
- –
- Introduced
- –
Parquet goes to CSV when data leaves the engineering side: an export from BigQuery, a Hugging Face dataset or a data lake table has to reach an analyst, a finance team or a tool that only reads rows of text.
Parquet to CSV questions
How can I open a Parquet file without Python?
Convert it here and open the CSV. Excel, Numbers and Google Sheets cannot read .parquet by double-click, and the usual alternatives, such as pandas, Spark or the DuckDB command line, all assume you are comfortable installing and running code.
Why is the CSV so much larger than the Parquet file?
Parquet stores each column together, encodes repeated values once in a dictionary and compresses the result, usually with Snappy or Zstandard. CSV writes every value out in full as text on every row. A CSV several times the size of its Parquet source is normal, so leave room on disk and check the row count before opening it in Excel.
What happens to nested struct and list columns?
A struct is spread into extra columns with dotted names, such as address.city. A list stays in one cell as JSON text, because a CSV row cannot hold a different number of cells per record.
My Spark job wrote a folder of part files. Which do I upload?
All of them. Spark, Hive and similar engines split one dataset into files named like part-00000.parquet, each a complete Parquet file with the same schema. Upload up to 20 together and you get one CSV per part, which you can combine afterward.
Can I turn the CSV back into Parquet after editing?
Yes, with CSV to Parquet. Column types are inferred again from the text, so check that IDs with leading zeros and decimal amounts come back with the types your pipeline expects.
- CSV to Parquet ConverterConvert CSV to Apache Parquet with typed columns and compression, ready for pandas, DuckDB, Spark, BigQuery or Athena. No Python or Spark setup needed.
- Excel to CSV ConverterConvert Excel workbooks (XLS, XLSX, XLSM) to CSV with every sheet as its own file, comma or semicolon delimiters and UTF-8 that Excel reopens cleanly.
- Convert JSON to CSVConvert JSON or JSON Lines to CSV: one row per object, nested fields flattened into columns, UTF-8 with BOM so Excel shows accents correctly. 20 files at once.
- CSV to XLSX ConverterConvert CSV to an Excel XLSX workbook with leading zeros, long IDs and accented text intact. Comma, semicolon and tab files detected automatically.
- CSV to SQL ConverterConvert CSV to SQL: a CREATE TABLE plus INSERT statements for MySQL, PostgreSQL, SQLite or SQL Server, with the table name you choose. No account.