The Tohoku Medical Megabank Project (TMM) was established to realize personalized healthcare and medicine using genomic and omics data. This study evaluated the validity and reproducibility of food group intakes derived from a self-administered food frequency questionnaire (TMM-FFQ) that included the response option "constitutionally unable to eat/drink it" among community-dwelling Japanese adults. Participants comprised 89 men and 124 women aged ≥20 years from Miyagi Prefecture. Participants completed weighed food records (WFRs) for 3 consecutive days per season as reference intake and FFQs in 2019 (FFQ1) and 2021 (FFQ3). Spearman's rank correlation coefficients (CCs) were calculated for correlations between food group intakes estimated from the 12-day WFR and FFQ3 (validity), and for correlations between those estimated from the FFQ1 and FFQ3 (reproducibility). Cross-classification according to quintiles using FFQ and WFR data was also performed. The percentage of participants who chose the "constitutionally unable to eat/drink it" option was non-negligible for some food groups. In the validity analysis, CCs were >0.40 for many food groups; the median across 21 food groups was 0.49 in men and 0.45 in women. The median percentages of cross-classification into exact plus adjacent quintiles were 73.0% in men and 66.9% in women. In the reproducibility analysis, CCs were >0.50 for many food groups; the median across 21 food groups was 0.60 in men and 0.51 in women. The validity of the TMM-FFQ compared with 12-day WFR and the reproducibility of the TMM-FFQ were reasonable for food groups in the TMM cohort studies.