Converting .sas7bdat to CSV Without SAS

.sas7bdat is the binary format SAS uses to store data sets. SAS is commercial statistical software with a licensing cost that prices out most individual users. The format is proprietary and undocumented by SAS, so every reader outside it, including pyreadstat, is the product of reverse engineering.

A single .sas7bdat file holds the table itself, the column metadata, and enough type information to reconstruct rows. What it does not hold is the display layer. SAS stores human-readable value labels and formats in a separate catalog file, .sas7bcat. Converting to CSV throws that layer away, because CSV has no place to put it.

Converting with pyreadstat

pyreadstat is a Python binding around ReadStat, a reverse-engineered C parser by Evan Miller. It reads the data set and reattaches value labels from the catalog.

import pyreadstat

df, meta = pyreadstat.read_sas7bdat(
    'input.sas7bdat',
    catalog_file='formats.sas7bcat',
    apply_value_formats=True,
)
df.to_csv('output.csv', index=False)

read_sas7bdat returns the DataFrame and a meta object. With apply_value_formats=True, a column coded 1=Yes, 2=No arrives in the CSV as Yes and No rather than 1 and 2. meta holds the variable labels and format definitions, which you can inspect or write to a sidecar file to keep alongside the CSV.

pyreadstat also decodes SAS dates correctly. SAS numbers days from 1960-01-01, not the 1970 Unix epoch. ReadStat applies that offset internally, so date columns arrive as proper datetimes.

Encoding

Older .sas7bdat files from Windows SAS are often wlatin1, not UTF-8. If you see mojibake in character columns, pass the encoding explicitly.

df, meta = pyreadstat.read_sas7bdat(
    'input.sas7bdat', encoding='latin-1',
    catalog_file='formats.sas7bcat', apply_value_formats=True,
)

Guessing wrong leaves the data intact but garbles accented characters. There is no penalty for trying a different encoding.

Beyond Python

R users reach the same result with haven, which also wraps ReadStat, so the two agree on output. For batch conversion of many files, sas7bdat-converter wraps a reader behind a command-line interface.

What the conversion loses

CSV cannot represent value labels, column formats, or SAS missing-value sentinels (., .A through .Z). A conversion is lossy by design, even with pyreadstat applying labels. If the missing-value distinctions matter for your analysis, keep the original .sas7bdat and the catalog, and treat the CSV as a transport format rather than an archive.