A locally validated tool was needed to evaluate long-term dietary intake in rural Bangladesh. We assessed the validity of a 42-item dish-based semi-quantitative food frequency questionnaire (FFQ) using two 3-day food diaries (FDs). We selected a random subset of 47 families (190 participants) from a longitudinal arsenic biomonitoring study in Bangladesh to administer the FFQ. Two 3-day FDs were completed by the female head of the households and we used an adult male equivalent method to estimate the FD for the other participants. Food and nutrient intakes measured by FFQ and FD were compared using Pearson's and Spearman's correlation, paired t-test, percent difference, cross-classification, weighted Kappa, and Bland-Altman analysis. Results showed good validity for total energy intake (paired t-test, p < 0.05; percent difference <10%), with no presence of proportional bias (Bland-Altman correlation, p > 0.05). After energy-adjustment and de-attenuation for within-person variation, macronutrient intakes had excellent correlations ranging from 0.55 to 0.70. Validity for micronutrients was mixed. High intraclass correlation coefficients (ICCs) were found for most nutrients between the two seasons, except vitamin A. This dish-based FFQ provided adequate validity to assess and rank long-term dietary intake in rural Bangladesh for most food groups and nutrients, and should be useful for studying dietary-disease relationships.