We conducted a population-based cohort study to determine the prevalence of antenatal and intra-partum referrals, compliance with advice and perinatal outcomes in referred pregnant women in Gutu district, Zimbabwe. The cohort was composed of 10,572 women who received antenatal care in 23 rural health centres (RHC) in Gutu district between January 1995 and June 1998. Pregnancy records of women with antenatal or intra-partum referral were analysed for indication, compliance and perinatal outcomes. Using women who had no antenatal referral or those who complied as referents, the association of referral with perinatal outcome was expressed as relative risk (RR) with 95% confidence intervals (CI). A total of 30% of women (3,094/10,572) had an antenatal referral. Among women attending RHC in labour, 13% (694/5,338) were referred intra-partum. Nulliparous and women younger than 20 years were more likely to be referred. Nurse - midwives' compliance with referral recommendations was low as 59% women with historical risk factors and 52% with raised blood pressure (>140/90 mmHg) were not referred. Women complied with referral advice except when indication was high parity. Women with antenatal referral were more likely to have hospital delivery, 70% vs 18% (p < 0.001). A total of 13% (993/7,478) of women referred themselves for hospital delivery. The risk of perinatal death was elevated among intra-partum referrals (RR 3.4; 95% CI 1.7 - 6.8), self-referrals (RR 2.6; 95% CI 1.5 - 4.5) and also among women with historical risk factors who were not referred (RR 4.8; 95% CI 2.5 - 9.2). We concluded that although there was a functional referral system in Gutu district its efficiency was reduced by failure of health personnel to comply with referral recommendations. Women took appropriate action for most referral indications.