BACKGROUND: To obtain a probability sample of pregnancies, the National Children's Study conducted door-to-door recruitment in randomly selected neighbourhoods in randomly selected counties in 2009-10. In 2011, an experiment was conducted in 10 US counties, in which the two-stage geographic sample was maintained, but participants were recruited in prenatal care provider offices. We describe our experience recruiting pregnant women this way in Wayne County, Michigan, a county where geographically eligible women attended 147 prenatal care settings, and comprised just 2% of total county pregnancies. METHODS: After screening for address eligibility in prenatal care offices, we used a three-part recruitment process: (1) providers obtained permission for us to contact eligible patients, (2) clinical research staff described the study to women in clinical settings, and (3) survey research staff visited the home to consent and interview eligible women. RESULTS: We screened 34,065 addresses in 67 provider settings to find 215 eligible women. Providers obtained permission for research contact from 81.4% of eligible women, of whom 92.5% agreed to a home visit. All home-visited women consented, giving a net enrolment of 75%. From birth certificates, we estimate that 30% of eligible county pregnancies were enrolled, reaching 40-50% in the final recruitment months. CONCLUSIONS: We recruited a high fraction of pregnancies identified in a broad cross-section of provider offices. Nonetheless, because of time and resource constraints, we could enrol only a fraction of geographically eligible pregnancies. Our experience suggests that the probability sampling of pregnancies for research could be more efficiently achieved through sampling of providers rather than households. 2013 Blackwell Publishing Ltd.