The reproducibility and validity of responses for 55 specific foods and beverages on a self-administered food frequency questionnaire were evaluated. One hundred and seventy three women from the Nurses' Health Study completed the questionnaire twice approximately 12 months apart and also recorded their food consumption for seven consecutive days, four times during the one-year interval. For the 55 foods, the mean of correlation coefficients between frequencies of intake for first versus second questionnaire was 0.57 (range = 0.24 for fruit punch to 0.93 for beer). The mean of correlation coefficients between the dietary records and first questionnaire was 0.44 (range = 0.09 for yellow squash to 0.83 for beer and tea) and between the dietary records and the second questionnaire was 0.52 (range = 0.08 for spinach to 0.90 for tea). Ratios of within- to between-person variance for the 55 foods were computed using the mean four one-week dietary records for each person as replicate measurements. For most foods this ratio was greater than 1.0 (geometric mean of ratios = 1.88), ranging from 0.25 (skimmed milk) to 14.76 (spinach). Correlation coefficients comparing questionnaire and dietary record for the 55 foods were corrected for the within-person variation (mean corrected value = 0.55 for dietary record versus first questionnaire and 0.66 versus the second). Mean daily amounts of each food calculated by the questionnaire and by the dietary record were also compared; the observed differences suggested that responses to the questionnaire tended to over-represent socially desirable foods. This analysis documents the validity and reproducibility of the questionnaire for measuring specific foods and beverages, as well as the large within-person variation for food intake measured by dietary records. Differences in the degree of validity for specific foods revealed in this type of analysis can be useful in improving questionnaire design and in interpreting findings from epidemiological studies that use the instrument.