This study evaluated the reproducibility and validity of the self-administered semiquantitative food frequency questionnaire used in a large prospective cohort study(Korean Cancer Research Survey) in middle-aged men. The questionnaire was administered twice at an interval of approximately two years(December, 1992-January, 1995), and four or five 24-hour recalls for each subject were collected at intervals of approximately three months. The results were as follows; 1) Although the distributions of the data estimated by the questionnaire were somewhat wider, the mean nutrient intakes of group estimated by our questionnaires and the multiple 24-hour recalls were roughly comparable. 2) The reproducibility determined by correlation of absolute(unadjusted nutrient intake) and calorie adjusted nutrient intakes from two semiquantitative food frequency questionnaires were more than 0.5, and the weighted kappa values were more than 0.4. 3) The Pearson correlation coefficients between unadjusted nutrient intakes values were average 0.40 on the average(Ca, 0.13-Carbohydrate, 0.58) at the first questionnaire vs. 24-hour recalls, and 0.28 at the second questionnaire vs. 24-hour recalls. The spearman rank order correlation coefficients were similar. When energy intake was adjusted, there was a slight reduction : 0.28 at the second questionnaire, 0.25 average on the second. In order to correct the measurement error of 24-hour recall data, the deattenuated correlation coefficient was calculated. It averaged 0.53 on the first questionnaire, 0.37 on the second questionnaire for unadjusted nutrient intake. for calorie-adjusted nutrient intake, it averaged 0.44 on the first questionnaire, 0.37 on the second questionnarie. 4) There was lower agreement(k<0.4) between the questionnaries and the 24-hour recalls. And the subjects classified in the same quartile by 24-hour recalls and first questionnaire were average 37$\%$(energy-adjusted values) and 40$\%$(unadjusted values) on the average. More than k10$\%$(average) of subjects were in the extreme quartile of the questionnarie and 24-hour recall method. But 8.2$\%$(average) of subjects classified in the lowest quartile of unadjusted nutrient intake level by the 24-hour recalls were in the highest quartile by the first questionnaire. These data indicate that our self-administered semiquantitative food frequency questionnarie is reproducible. Correlation coefficients comparing nutrient intakes measured by two different dietary assessment methods were less than 0.5. The validity of our questionnarie is not high enough.