How to Open a Parquet File
Someone sent you a .parquet file and Excel refuses to touch it. Parquet is a columnar format built for data pipelines, not office software — but opening one is easier than it looks.
What a Parquet file actually is
Apache Parquet stores data by column rather than by row, with the schema and statistics written in a footer at the end of the file. That design makes it extremely efficient for analytics — a query that needs two columns of a hundred reads only those two — and it is why Parquet is the default format of data lakes, Spark jobs and warehouse exports. It also means there is no plain-text inside to peek at: you need a real Parquet reader.
Option 1 — Open it in your browser (fastest)
If you just need to see what is inside — the columns, their types, and the data — use the Parquet Viewer on this site. Drop the file in; it reads the footer metadata and rows locally in your browser, shows the schema and a data preview, and can export everything to CSV. Nothing is installed and the file is never uploaded, which matters when the data is sensitive. Browser memory is the limit: comfortable up to roughly 100,000 rows for full reads, while the schema always loads instantly.
Option 2 — Python with pandas
The data-team standard. With Python installed:
pip install pandas pyarrow
import pandas as pd
df = pd.read_parquet("data.parquet")
print(df.head())
pyarrow is the engine doing the reading; pandas gives you the DataFrame. Best when you will go on to analyse the data, and overkill when you only needed a look inside.
Option 3 — DuckDB
DuckDB reads Parquet natively and lets you query it with SQL without loading everything into a DataFrame:
SELECT * FROM 'data.parquet' LIMIT 10; DESCRIBE SELECT * FROM 'data.parquet';
The second command is the quickest way to inspect a schema from the command line. Ideal for large files you want to filter before exporting.
Option 4 — Spark or your warehouse
If the file came from a data platform, the platform itself is a reader: Spark's spark.read.parquet(...), or the external-table features of BigQuery, Snowflake and Athena. Sensible when the file is one of thousands in a lake; unnecessary for a single file someone emailed you.
Which should you choose?
Just looking, or converting to CSV → the browser viewer. Ongoing analysis in Python → pandas + pyarrow. SQL against a big file → DuckDB. Part of a pipeline → the platform that produced it. And if the file will not open anywhere, check it is really Parquet: genuine files start and end with the magic bytes PAR1 — a renamed CSV or a truncated download fails every reader for the same reason.