How to upload data to BigQuery: CSV, Google Drive and SQL

Before you start

• A Google Cloud project with BigQuery enabled, and a dataset to hold the table.
• Permission to create tables in it. The BigQuery Data Editor role includes that.
• The file in a supported format: CSV, JSON (newline-delimited), Avro, Parquet or ORC.

Upload a CSV from your computer

1. In the BigQuery console, find your dataset in the Explorer pane, open its menu and choose Create table.
2. Under Create table from, choose Upload and select the file.
3. Set File format to CSV.
4. Give the table a name.
5. Under Schema, tick Auto detect, or type the column names and types yourself.
6. Open Advanced options and set Header rows to skip to 1 if the first row holds column names.
7. Click Create table.

The console creates a load job. When it finishes, the table appears under the dataset.

Two limits apply to local uploads. A file cannot exceed 100 MB in the console, and files must be loaded one at a time: no wildcards, no lists.

Larger files: Cloud Storage and LOAD DATA

Upload the file to a Cloud Storage bucket, then load it with SQL:

LOAD DATA INTO mydataset.orders
FROM FILES (
  format = 'CSV',
  uris = ['gs://my-bucket/orders/2026-10-*.csv'],
  skip_leading_rows = 1
);

LOAD DATA INTO appends to the table and creates it if it does not exist. LOAD DATA OVERWRITE replaces its contents. From Cloud Storage you can use a wildcard, as above, to load many files in one statement. To fix the schema instead of detecting it, list the columns after the table name:

LOAD DATA OVERWRITE mydataset.orders
(order_id INT64, customer STRING,
 amount NUMERIC, order_date DATE)
FROM FILES (
  format = 'CSV',
  uris = ['gs://my-bucket/orders/orders.csv'],
  skip_leading_rows = 1
);

Both statements follow Google’s reference syntax. They were not run for this article, since they write to a dataset.

The same load can be done from the console by choosing Google Cloud Storage as the source in the Create table dialog, or with the bq load command.

Batch loading is free. You are charged for the storage of the loaded data, and for the file while it sits in Cloud Storage, so delete it there once the load is checked.

From Google Drive or Google Sheets

For a file in Drive, BigQuery can read it in place through an external table:

1. In Create table, choose Drive as the source.
2. Paste the file’s URL.
3. Choose the format: CSV, JSON, Avro or Google Sheet. For a sheet you can name the tab and range.
4. Set the schema and create the table.

Nothing is copied. Each query reads the file as it is at that moment, so edits in the sheet show up immediately, and so do mistakes. The account that runs the query needs access to the Drive file. For data a report depends on, materialize the external table into a normal one:

CREATE OR REPLACE TABLE mydataset.targets AS
SELECT * FROM mydataset.targets_sheet;

More on this route in how to connect Google Sheets to BigQuery.

Schema: detect it or declare it

Auto detect reads a sample of the file and guesses each column’s type. It is fine for a first look. For anything recurring, declare the schema, because detection can change its mind between files: a column of IDs that happen to be numeric this week becomes INT64, and next week’s file with one letter in it fails.

One documented quirk: if every column in a CSV is text, auto detect cannot tell the header row from the data. Declare the schema or add a numeric column.

Errors that stop a load

• Dates in the wrong layout. In CSV and JSON, a DATE column must be YYYY-MM-DD. Load 06/10/2026 as STRING and convert it with PARSE_DATE.
• Thousands separators. 1,200 is not a number to a load job. Load as STRING and use PARSE_NUMERIC.
• A byte order mark at the start of the file. Remove it; Google’s page says it can cause unexpected issues.
• Nested data in CSV. CSV cannot hold arrays or structs. Use JSON, Avro or Parquet.
• Compressed files. A gzip CSV loads, but not in parallel, and cannot exceed 4 GB. Compressed and uncompressed files cannot be mixed in one job.
• Too many bad rows. A load fails on the first bad row unless you allow some under Advanced options. Allowing them hides the problem, so count what was skipped.

A sound pattern for messy files: load everything as STRING into a raw table, then build the typed table with a query that uses SAFE_CAST and counts the failures. See CAST and SAFE_CAST.

When uploading is the wrong tool

A manual upload is right for a one-off file: a list of targets, a mapping table, an export someone sent you. It is the wrong habit for anything that repeats. The third time you upload this week’s CSV, the task wants a scheduled load. The options, from the Data Transfer Service to connectors for ad platforms, are compared in how to load data into BigQuery.

For ad platforms and similar sources, OWOX Data Marts has open-source connectors that load into your warehouse on a schedule, and it keeps the queries you write over that data, with their owners and run history, in the same place. OWOX does not clean or reshape the files for you; the SQL stays yours.

Google BigQueryHow to connect BigQuery to Google Sheets: three waysOctober 6, 2026→Google BigQueryBigQuery Sandbox 2025: Beginner's Setup and TipsOctober 3, 2024→Google BigQueryComprehensive Guide to BigQuery Aggregate Functions for 2025April 18, 2024→

See all articles →