The birth certificate, the primary tool for population-based surveillance of the condition of infants at birth and maternal status during pregnancy, provides little data about maternal behaviour during pregnancy. To collect data on maternal behaviours that influence pregnancy outcome, we implemented the Pregnancy Risk Assessment Monitoring System in seven states. For this population-based surveillance, new mothers were sampled from birth certificates 2 to 6 months after delivery and contacted by mail; follow-up of nonrespondents was by telephone. Participants completed a 10-page questionnaire. Stratification permitted over-sampling of women with adverse pregnancy outcomes. Among 10,563 women sampled during 1988 and 1989, stratum-specific response rates ranged from 30% to 89%. In 11 of the 28 strata, response rates were greater than 70%. Response rates varied considerably between states. Rates were lower for Black mothers, mothers of low birthweight infants, unmarried mothers and mothers with less than 12 years of education. Active refusal to participate and undelivered mail occurred infrequently. Mail and telephone surveillance of new mothers can yield adequate response rates in selected population groups. Trials of alternative approaches to enhancing response among Black and disadvantaged mothers, such as additional mailings or post-partum in-hospital recruitment, are needed.