The aim of this study was to develop a self-administered food frequency questionnaire for use with low-income pregnant women and to evaluate its performance in classifying women according to nutrient intake. Index nutrients used were energy, protein, calcium, iron, zinc, and vitamins A, B-6, and C. Two hundred ninety-five Massachusetts women, aged 14 to 43 years, participated in the field test of the questionnaire. A subset of 95 women provided three 24-hour diet recalls for use in comparative studies. Correlation coefficients between questionnaire and diet recall scores were adjusted for measurement error resulting from the limited number of 24-hour recalls per subject, and their confidence intervals were computed. When subjects with implausibly high energy scores (greater than 4,500/day) were removed from the sample, reducing sample size by about 15%, correlation coefficients increased substantially (25% to 64%) for all nutrients except vitamin A. Adjusted correlation coefficients exceeded 0.5, excluding vitamin A (r approximately 0.15), and quintile comparisons indicated that the questionnaire would correctly identify a high proportion of the women having low intake of selected nutrients. We conclude that a self-administered questionnaire can provide useful data about individual recent intake of selected nutrients in a majority of English-speaking, low-income pregnant women, but that overestimation of food use may occur among up to 20% of this population.