An antenatal HIV testing programme was introduced in June 1995. All women at their booking visit were offered an HIV test by midwives trained about HIV infection and how to discuss testing. Each midwife was allocated a number (by a senior midwife not involved in the study) and gave details of age, year of qualification, self defined ethnic group, and site of work. For each pregnant woman the midwife's number, duration of pretest discussion, decision about testing, and whether the result was to be posted or collected were recorded. Only negative results were sent by post. Uptake was calculated as the percentage of women booked who accepted an HIV test. Logistic regression models for grouped data (STATA version 4.0, Austin, TX, USA) were used to estimate odds ratios.