BEDCA Food Composition Dataset

Food composition data from the Spanish Food Composition Database, retrieved from https://www.bedca.net/ (the database app at https://www.bedca.net/bdpub/) in August 2026 via the site's public XML query interface, at a polite rate (one sequential request every 2.5 s).

Download

• GitHub release — bedca-dataset-2026.zip: the complete dataset (README, LICENSE, output/ CSV files, manifest). Verify with the published SHA-256 checksum in bedca-dataset-2026.zip.sha256: sha256sum -c bedca-dataset-2026.zip.sha256
• Zenodo archive — https://doi.org/10.5281/zenodo.22017000: a citable, versioned record of this repository. The concept DOI 10.5281/zenodo.22017001 always resolves to the latest version.

Attribution

Source: AESAN/BEDCA Base de Datos Española de Composición de Alimentos v1.0 (2010)

This attribution must accompany any reproduction or publication of this data, including in application credits.

Terms of use (summary)

Per the AESAN/BEDCA conditions of use (mirrored in LICENSE, original in terms-of-use-txt):

• The data may be reproduced only with clear indication of the original source (see the attribution string above).
• The data may not be modified, nor may its original meaning be altered. All values in this dataset are therefore stored verbatim, exactly as returned by the database (no rounding, no unit conversion, no normalization).
• Use is limited to personal, educational, or non-commercial purposes. Reproduction, translation, or any other use of the data in electronic format requires the express authorization of AESAN/BEDCA.
• Errors in the data should be reported to bedca.adm@gmail.com.

Contents (output/)

File	Format	Description
bedca_foods.csv	wide	1 row per food. Food metadata, then one value + one unit column per nutrient component (e.g. ENERC, ENERC_unit). Values verbatim. UTF-8 with BOM, CRLF.
bedca_foodvalues.csv	long	1 row per (food, component) — the full-fidelity record, including per-value stdv, min, max, v_n, scientific citation/ref_id, and method provenance.
bedca_units.csv	long	Distinct component/unit combinations (v_unit, u_descripcion, mu_descripcion) — the unit reference for downstream merging.
bedca_index.csv	long	Food list snapshot: f_id, Spanish/English names, LanguaL codes, f_origen, edible_portion.
manifest.json	JSON	Fetch log per food (status, bytes, retries), failures, and the parse report.

Field definitions come from the EuroFIR documentation referenced by AESAN/BEDCA; key ones: best_location = the reported value, v_unit = unit of the value, moex = method of expression (W = per 100 g of edible portion), value_type = AR as reported / BE best estimate, edible_portion = edible fraction as stored by the source database.

Usage

The CSV files are UTF-8 with BOM + CRLF. Load them in pandas with encoding="utf-8-sig" — this strips the BOM and reads Spanish diacritics correctly:

import pandas as pd

foods = pd.read_csv("output/bedca_foods.csv", encoding="utf-8-sig")
foodvalues = pd.read_csv("output/bedca_foodvalues.csv", encoding="utf-8-sig")

bedca_foods.csv has one row per food (one value + one unit column per nutrient); bedca_foodvalues.csv has one row per measured component with full provenance. See the table above for column details.

Known data characteristics

• The public database returned by the site's list query contains 969 foods (public subset, publico=1), more than the commonly cited ~500.
• 957 foods have composition data (31,166 component values); 12 foods are listed in the index but return an empty <food/> element from the detail endpoint — they have no public data. Their ids are recorded in manifest.json → empty (e.g. 2412, 2509, 2702–2710, 2714).
• 47 of the catalog's components occur in the data; per-food coverage varies (roughly 30 components per food), and a food may carry a component row with an empty value.
• f_origen contains both BEDCA and BEDCA2 values — a property of the source database's own filtering, reproduced verbatim.
• Five components appear with more than one unit across foods (BIOT, CHO, ENERC, FIBT, WATER; see manifest.json → parse.multi_unit_components). Values and units are stored as-is; downstream merging logic decides how to handle that.