data.table fread() in R: Fast CSV and Text Import

The fread() function from data.table imports CSV and delimited text files into R quickly, auto-detecting the separator, header, and column types so you rarely need to configure anything.

fread("data.csv")                          # auto-detect sep, header, types
fread("data.tsv")                          # tab-delimited, still automatic
fread(text = "a,b\n1,2")                   # read straight from a string
fread("data.csv", select = c("a", "b"))    # keep only these columns
fread("data.csv", drop = "notes")          # read all columns except these
fread("data.csv", nrows = 100)             # read just the first 100 rows
fread("data.csv", colClasses = c(id = "character"))  # force a column type

What fread() does in one sentence

fread() is data.table's fast file reader. You give it a path to a CSV or other delimited file, and it returns a data.table with the separator, header row, and column types all inferred automatically. The name stands for "fast read", and on large files it is typically several times quicker than base R read.csv() while using less memory.

The appeal of data.table fread is that it usually just works. It detects whether your file is comma, tab, semicolon, or pipe separated. It recognises integers, doubles, dates, and logicals without you listing them. It handles quoted fields, embedded newlines, and irregular whitespace. For most day-to-day imports, fread("myfile.csv") is the entire command.

Syntax

fread() takes a source plus a set of optional controls. The source is usually a file path, but it can also be a URL, a shell command, or a literal string passed through the text argument.

The most-used arguments are:

• input: a file path, URL, or connection. The first positional argument.
• text: a character string to read instead of a file.
• sep: the field separator. Default "auto" detects it for you.
• header: whether row one holds column names. Default "auto".
• select / drop: columns to keep or skip.
• nrows: maximum number of rows to read.
• colClasses: force specific column types.
• na.strings: strings to treat as NA.

Reading data: five common patterns

1. Read from a string with text

When data is already in your script, or comes back from an API as text, skip the file entirely and pass it through text.

Note that the separator here is a semicolon, and fread() detected it without being told.

2. Keep only some columns with select

Reading just the columns you need is faster and lighter than reading everything and dropping later.

3. Drop columns you do not want

drop is the inverse of select: read every column except the ones you name.

4. Preview a large file with nrows

Before loading a multi-gigabyte file, read a handful of rows to check the structure.

5. Force column types with colClasses

Type detection is good but not psychic. When you need a specific type, name it.

fread() vs read.csv() and read_csv()

The three readers differ in speed, return type, and defaults. fread() is fastest and returns a data.table. Base read.csv() returns a data.frame. readr::read_csv() returns a tibble and sits between the two on speed.

Use fread() for large files or when you already work in data.table. Use read_csv() if the rest of your code is tidyverse. Reach for read.csv() only when you want zero package dependencies.

Common pitfalls

Pitfall 1: leading-zero codes become integers. fread() sees a column of digits and reads it as a number, dropping leading zeros from ZIP codes or IDs.

Pitfall 2: a string with no newline is treated as a filename. fread("a,b,c") looks for a file named a,b,c. To read literal text, always use the text argument.

Pitfall 3: forgetting the result is a data.table. Indexing such as dt[1] returns the first row, not the first column, which surprises users expecting data.frame behaviour.

Try it yourself

Related data.table functions

Once fread() has loaded your data, these functions pair naturally with it:

• fwrite(): the fast counterpart that writes a data.table back to disk.
• setDT(): convert an existing data.frame to a data.table in place.
• rbindlist(): stack a list of data.tables, ideal for combining many files read with fread().
• setnames(): rename columns of the result.
• setkey(): index a column for fast joins and lookups.

A common pattern combines several of these: rbindlist(lapply(list.files(pattern = "*.csv"), fread)) reads every CSV in a folder and binds them into one data.table in a single line.

FAQ

Is fread faster than read.csv?

Yes, substantially. On files above 100 MB, fread() is commonly 5 to 10 times faster than read.csv(), and the gap widens as files grow. It achieves this with a memory-mapped, multi-threaded parser and by sampling the file to guess column types once instead of growing vectors row by row. On tiny files the difference is negligible, so the win is mainly for medium and large data.

How do I read only specific columns with fread?

Use the select argument with a character vector of column names: fread("data.csv", select = c("id", "price")). You can also pass column positions, such as select = c(1, 3). The opposite is drop, which reads every column except the ones you name. Selecting columns at read time is faster and uses less memory than reading the whole file first.

Does fread return a data.frame or a data.table?

fread() returns a data.table, which also inherits from data.frame. That means base R functions expecting a data frame still work, but you also get data.table's fast dt[i, j, by] query syntax. If you need a plain data frame, set data.table = FALSE in the call or wrap the result with as.data.frame().

How do I read a CSV from a URL with fread?

Pass the URL as the first argument: fread("https://example.com/data.csv"). fread() downloads the file and parses it in one step, with the same automatic detection it uses for local files. For files behind authentication or needing custom headers, download the file first, then read the local copy.

Why does fread change my column types?

fread() infers each column's type from a sample of values. A column of digits becomes integer or numeric, and TRUE/FALSE text becomes logical. This is usually helpful, but it can drop leading zeros or misread mixed columns. Override it with colClasses, for example colClasses = c(zip = "character"), to lock a column to the type you need.

For the official argument reference, see the data.table fread documentation.