Read in a data.frame from a file. Exceptions to this rule are Rdata, RDS, and JSON input file formats, which return the originally saved object without changing its class.

import(file, format, setclass, which, ...)



A character string naming a file, URL, or single-file .zip or .tar archive.


An optional character string code of file format, which can be used to override the format inferred from file. Shortcuts include: “,” (for comma-separated values), “;” (for semicolon-separated values), and “|” (for pipe-separated values).


An optional character vector specifying one or more classes to set on the import. By default, the return object is always a “data.frame”. Allowed values include “tbl_df”, “tbl”, or “tibble” (if using dplyr) or “data.table” (if using data.table). Other values are ignored, such that a data.frame is returned.


This argument is used to control import from multi-object files; as a rule import only ever returns a single data frame (use import_list to import multiple data frames from a multi-object file). If file is a compressed directory, which can be either a character string specifying a filename or an integer specifying which file (in locale sort order) to extract from the compressed directory. For Excel spreadsheets, this can be used to specify a sheet name or number. For .Rdata files, this can be an object name. For HTML files, it identifies which table to extract (from document order). Ignored otherwise. A character string value will be used as a regular expression, such that the extracted file is the first match of the regular expression against the file names in the archive.


Additional arguments passed to the underlying import functions. For example, this can control column classes for delimited file types, or control the use of haven for Stata and SPSS or readxl for Excel (.xlsx) format. See details below.


A data frame. If setclass is used, this data frame may have additional class attribute values, such as “tibble” or “data.table”.


This function imports a data frame or matrix from a data file with the file format based on the file extension (or the manually specified format, if format is specified).

import supports the following file formats:

  • Comma-separated data (.csv), using fread or, if fread = FALSE, read.table with row.names = FALSE and stringsAsFactors = FALSE

  • Pipe-separated data (.psv), using fread or, if fread = FALSE, read.table with sep = '|', row.names = FALSE and stringsAsFactors = FALSE

  • Tab-separated data (.tsv), using fread or, if fread = FALSE, read.table with row.names = FALSE and stringsAsFactors = FALSE

  • SAS (.sas7bdat), using read_sas.

  • SAS XPORT (.xpt), using read_xpt or, if haven = FALSE, read.xport.

  • SPSS (.sav), using read_sav. If haven = FALSE, read.spss can be used.

  • SPSS compressed (.zsav), using read_sav.

  • Stata (.dta), using read_dta. If haven = FALSE, read.dta can be used.

  • SPSS Portable Files (.por), using read_por.

  • Excel (.xls and .xlsx), using read_excel. Use which to specify a sheet number. For .xlsx files, it is possible to set readxl = FALSE, so that read.xlsx can be used instead of readxl (the default).

  • R syntax object (.R), using dget

  • Saved R objects (.RData,.rda), using load for single-object .Rdata files. Use which to specify an object name for multi-object .Rdata files. This can be any R object (not just a data frame).

  • Serialized R objects (.rds), using readRDS. This can be any R object (not just a data frame).

  • Epiinfo (.rec), using read.epiinfo

  • Minitab (.mtp), using read.mtp

  • Systat (.syd), using read.systat

  • "XBASE" database files (.dbf), using read.dbf

  • Weka Attribute-Relation File Format (.arff), using read.arff

  • Data Interchange Format (.dif), using read.DIF

  • Fortran data (no recognized extension), using read.fortran

  • Fixed-width format data (.fwf), using a faster version of read.fwf that requires a widths argument and by default in rio has stringsAsFactors = FALSE. If readr = TRUE, import will be performed using read_fwf, where widths should be: NULL, a vector of column widths, or the output of fwf_empty, fwf_widths, or fwf_positions.

  • gzip comma-separated data (.csv.gz), using read.table with row.names = FALSE and stringsAsFactors = FALSE

  • CSVY (CSV with a YAML metadata header) using fread.

  • Apache Arrow Parquet (.parquet), using read_parquet

  • Feather R/Python interchange format (.feather), using read_feather

  • Fast storage (.fst), using read.fst

  • JSON (.json), using fromJSON

  • Matlab (.mat), using read.mat

  • EViews (.wf1), using readEViews

  • OpenDocument Spreadsheet (.ods), using read_ods. Use which to specify a sheet number.

  • Single-table HTML documents (.html), using read_html. The data structure will only be read correctly if the HTML file can be converted to a list via as_list.

  • Shallow XML documents (.xml), using read_xml. The data structure will only be read correctly if the XML file can be converted to a list via as_list.

  • YAML (.yml), using yaml.load

  • Clipboard import (on Windows and Mac OS), using read.table with row.names = FALSE

  • Google Sheets, as Comma-separated data (.csv)

  • GraphPad Prism (.pzfx) using read_pzfx

import attempts to standardize the return value from the various import functions to the extent possible, thus providing a uniform data structure regardless of what import package or function is used. It achieves this by storing any optional variable-related attributes at the variable level (i.e., an attribute for mtcars$mpg is stored in attributes(mtcars$mpg) rather than attributes(mtcars)). If you would prefer these attributes to be stored at the data.frame-level (i.e., in attributes(mtcars)), see gather_attrs.

After importing metadata-rich file formats (e.g., from Stata or SPSS), it may be helpful to recode labelled variables to character or factor using characterize or factorize respectively.


For csv and txt files with row names exported from export, it may be helpful to specify row.names as the column of the table which contain row names. See example below.

See also


# create CSV to import export(iris, "iris1.csv") # specify `format` to override default format export(iris, "iris.tsv", format = "csv") stopifnot(identical(import("iris1.csv"), import("iris.tsv", format = "csv"))) # import CSV as a `data.table` stopifnot(inherits(import("iris1.csv", setclass = "data.table"), "data.table")) # pass arguments to underlying import function iris1 <- import("iris1.csv") identical(names(iris), names(iris1))
#> [1] TRUE
export(iris, "iris2.csv", col.names = FALSE) iris2 <- import("iris2.csv") identical(names(iris), names(iris2))
#> [1] FALSE
# set class for the response data.frame as "tbl_df" (from dplyr) stopifnot(inherits(import("iris1.csv", setclass = "tbl_df"), "tbl_df")) # non-data frame formats supported for RDS, Rdata, and JSON export(list(mtcars, iris), "list.rds") li <- import("list.rds") identical(names(mtcars), names(li[[1]]))
#> [1] TRUE
# cleanup unlink("iris.tsv") unlink("iris1.csv") unlink("iris2.csv") unlist("list.rds")
#> [1] "list.rds"