The population under study in the South African longitudinal study of urban children and their families, 'Birth to Ten' (BTT), comprised all births during a 7-week period from April to June 1990 in Soweto-Johannesburg. Specification of the population base for the cohort was hampered by a number of flaws in the notification and record-keeping systems of the local authorities. As far as could be ascertained, 5460 singleton births occurred during this time to women who gave a permanent address within the defined region. Enrolment into BTT took place over the first 15 months of the study and covered the antenatal, delivery, 6-month and 1-year periods. By the end of this time, and despite a major health service strike during the delivery phase, 74% of all births (4029 cases) had been enrolled into the study. There were marked variations in levels of enrolment, however, by population group membership, residential area and place of delivery. In general, there was substantial under-enrolment of largely middle-class white women and their babies. Initial non-enrolment of specific segments of the population and attrition of the enrolled sample up to the end of the first year are discussed in the context of racial and social differentiation in South Africa.