Background The inequity in emergency obstetric care access in Tanzania is unsatisfactory. Despite an existing national obstetric referral system, many birthing women bypass referring facilities and go directly to higher-level care centres. We wanted to compare Caesarean section (CS) rates among women formally referred to a tertiary care centre versus self-referred women, and to assess the effect of referral status on adverse outcomes after CS. Methods We used data from 21,011 deliveries, drawn from the birth registry of a tertiary hospital in northeastern Tanzania, during 2000-07. Referral status was categorized as self-referred if the woman had bypassed or not accessed referral, or formally-referred if referred by a health worker. Because CS indications were insufficiently registered, we applied the Ten-Group Classification System to determine the CS rate by obstetric group and referral status. Associations between referral status and adverse outcomes after CS delivery were analysed using multiple regression models. Outcome measures were CS, maternal death, obstetric haemorrhage ≥ 750 mL, postpartum stay > 9 days, neonatal death, Apgar score < 7 at 5 min and neonatal ward transfer. Results Referral status contributed substantially to the CS rate, which was 55.0% in formally-referred and 26.9% in self-referred birthing women. In both groups, term nulliparous singleton cephalic pregnancies and women with previous scar(s) constituted two thirds of CS deliveries. Low Apgar score (adjusted OR 1.42, 95% CI 1.09-1.86) and neonatal ward transfer (adjusted OR 1.18, 95% CI 1.04-1.35) were significantly associated with formal referral. Early neonatal death rates after CS were 1.6% in babies of formally-referred versus 1.2% in babies of self-referred birthing women, a non-significant difference after adjusting for confounding factors (adjusted OR 1.37, 95% CI 0.87-2.16). Absolute neonatal death rates were > 2% after CS in breech, multiple gestation and preterm deliveries in both referral groups. Conclusions Women referred for delivery had higher CS rates and poorer neonatal outcomes, suggesting that the formal referral system successfully identifies high-risk birth, although low volume suggests underutilization. High absolute rates of post-CS adverse outcomes among breech, multiple gestation and preterm deliveries suggest the need to target self-referred birthing women for earlier professional intrapartum care.