SETTING: Thyolo District Hospital, rural Malawi. OBJECTIVES: In a prevention of mother-to-child HIV transmission (PMTCT) programme, to determine: the acceptability of offering 'opt-out' voluntary counselling and HIV-testing (VCT); the progressive loss to follow up of HIV-positive mothers during the antenatal period, at delivery and to the 6-month postnatal visit; and the proportion of missed deliveries in the district. DESIGN: Cohort study. METHODS: Review of routine antenatal, VCT and PMTCT registers. RESULTS: Of 3136 new antenatal mothers, 2996 [96%, 95% confidence interval (CI): 95-97] were pre-test counselled, 2965 (95%, CI: 94-96) underwent HIV-testing, all of whom were post-test counselled. Thirty-one (1%) mothers refused HIV-testing. A total of 646 (22%) individuals were HIV-positive, and were included in the PMTCT programme. Two hundred and eighty-eight (45%) mothers and 222 (34%) babies received nevirapine. The cumulative loss to follow up (n=646) was 358 (55%, CI: 51-59) by the 36-week antenatal visit, 440 (68%, CI: 64-71) by delivery, 450 (70%, CI: 66-73) by the first postnatal visit and 524 (81%, CI: 78-84) by the 6-month postnatal visit. This left just 122 (19%, CI: 16-22) of the initial cohort still in the programme. The great majority (87%) of deliveries occurred at peripheral sites where PMTCT was not available. CONCLUSIONS: In a rural district hospital setting, at least 9 out of every 10 mothers attending antenatal services accepted VCT, of whom approximately one-quarter were HIV-positive and included in the PMTCT programme. The progressive loss to follow up of more than three-quarters of this cohort by the 6-month postnatal visit demands a 'different way of acting' if PMTCT is to be scaled up in our setting.