OBJECTIVE: To develop and validate a photographic food atlas of common foods for dietary assessment in southern Nepal. DESIGN: We created a life-sized photographic atlas of forty locally prepared foods. Between March and June 2014, data collectors weighed portion sizes that respondents consumed during one mealtime and then a different data collector revisited the household the next day to record respondents' estimations of their previous day's intakes using the atlas. Validity was assessed by percentage error, Cohen's weighted kappa (κ w) and Bland-Altman limits of agreement. SETTING: Dhanusha and Mahottari districts in southern Nepal. SUBJECTS: A random sample of ninety-five adults in forty-eight rural households with a pregnant woman. RESULTS: Overall, respondents underestimated their intakes (mean error =-4·5 %). Rice and dal (spiced lentil soup) intakes were underestimated (-14·1 % and -34·5 %, respectively), but vegetable curry intake was overestimated (+20·8 %). Rice and vegetable curry portion size images were significantly reliably selected (Cohen's κ w (se): rice=0·391 (0·105); vegetable curry=0·430 (0·139)), whereas dal images were not. Energy intake over one mealtime was under-reported by an average of 569 kJ (136 kcal; 4·5 % error) using recall compared with the weighing method. CONCLUSIONS: The photographic atlas is a useful tool for field estimation of dietary intake. Average errors were low, and there was 'modest' agreement between weighed and recalled portion size image selection of rice and vegetable curry food items. Error in energy estimation was low but with wide limits of agreement, suggesting that there is scope for future work to reduce error further.