This study tested the ability of a self-administered food frequency questionnaire (FFQ) to identify women with low fat intakes. FFQs were completed by 95 control participants of a dietary trial at a mean 2.9 +/- 0.8 years post-randomization. Subjects were selected in approximately equal numbers from women who were low-fat eaters (< or = 30% of energy from fat) and high-fat eaters (> 30% of energy from fat). Percentage energy from fat derived from food records and FFQ were similar in both the low- and high-fat eaters. Percentage of energy from carbohydrate and total grams of carbohydrate (low-fat eaters only) were slightly higher measured by FFQ than by food records, and percentage of energy from protein was slightly lower. The correlation between nutrient intake measured by FFQ and food records for the whole group was 0.74 for percentage of energy from fat, 0.50 for total fat, 0.59 for percentage of energy from carbohydrate, 0.43 for total carbohydrate, 0.53 for percentage of energy from protein, 0.27 for total protein, and 0.32 for energy intake. Correlations were slightly lower when the low- and high-fat eaters were examined separately. The area under the receiver operating characteristic (ROC) curve, 0.83, was significantly above 0.5 (p < < 0.001), indicating that the FFQ discriminated between low- and high-fat eaters significantly better than chance. The FFQ cutoff point of 30% of energy from fat had a true positive rate of 0.63 and false positive rate of 0.24. The use of this cutoff point for screening would result in the loss of 36% of potential subjects and an estimated increase in baseline percentage of energy from fat intake of 2.3 percentage points.