Input and output Updated November 2026

pd.read_csv

pd.read_csv(path) reads a CSV file (or URL) into a DataFrame, using the first row as column names and detecting number types automatically. Use sep for other delimiters, usecols to keep only some columns, parse_dates to turn date columns into datetimes and dtype to force column types.

Syntax

pd.read_csv(filepath_or_buffer, sep=",", header=0, names=None, index_col=None, usecols=None, dtype=None, parse_dates=None, nrows=None, encoding=None)

Part	What it is
`filepath_or_buffer`	Path, URL or file-like object to read
`sep`	Delimiter, default comma; use "\t" for tab files or ";" for European CSVs
`header`	Row number used as column names (0 by default); None if the file has no header
`index_col`	Column name or position to use as the row index
`usecols`	List of column names or positions to load; the rest is skipped
`parse_dates`	List of columns to convert to datetime while reading

Examples

import pandas as pd

df = pd.read_csv("sales.csv")
df.head(2)

Result: order_id date region amount 0 1001 2024-01-05 East 120.5 1 1002 2024-01-06 West 80.0

The first row becomes the header and amount is parsed as float automatically; date stays a string unless you parse it.

df = pd.read_csv("sales.csv", parse_dates=["date"], usecols=["date", "region", "amount"])
df.dtypes

Result: date datetime64[ns] region object amount float64 dtype: object

df = pd.read_csv("export.csv", sep=";", decimal=",", encoding="latin-1")

Typical settings for a CSV exported from Excel in a European locale.

Common mistakes

FileNotFoundError: the path is relative to the working directory of the Python process, not to the script, so print os.getcwd() or use an absolute path.
Everything lands in one column: the file is not comma separated, pass sep=";" or sep="\t" (or sep=None, engine="python" to sniff it).
UnicodeDecodeError: the file is not UTF-8, try encoding="latin-1" or encoding="cp1252".

Practice pd.read_csv

These drills check your answer on live data, free, no account needed.

Pick two columnsFrom the `emp` table, keep only the `name` and `salary` columns, in that order. The result is a DataFrame with every row.
Dates with to_datetime and dtThe `joined` column holds dates as text. Return a DataFrame with the columns `user`, `month` (the month number) and `weekday` (the full day name, like Monday).

Questions

How do I read a CSV with no header row?

Pass header=None so the first row is treated as data, and names=["a", "b", "c"] to give the columns names.

How do I read only some rows?

nrows=N reads the first N rows; skiprows=10 skips the first 10 lines; chunksize=100000 returns an iterator of DataFrames for files too large for memory.

Can read_csv read from a URL or a zip file?

Yes. Pass an http(s) URL directly, and a .zip or .gz path is decompressed automatically based on the extension.

to_csv and to_excelHow to export a pandas DataFrame with to_csv and to_excel: drop the index, choose separator and encoding, format floats, several sheets, append to a CSV.
pd.to_datetimeHow to parse dates in pandas with pd.to_datetime: formats, day-first dates, coercing bad values, and extracting year, month and weekday with the dt accessor.
astypeHow to change a column type in pandas with astype and pd.to_numeric: text to numbers, float to int, strings, categories, nullable Int64, bad values and NaN.
head, tail and sampleHow to look at a pandas DataFrame with head, tail and sample: the first or last N rows, random rows with a seed, a fraction of rows, and showing all rows.