Objective: Assess to what extent caesarean section (CS) indications followed evidence-based, locally co-created guidelines and identify reasons contributing to non-medically indicated CSs. Design: Retrospective cross-sectional study. Setting: Five urban, high-volume maternity units in Dar es Salaam, Tanzania. Population: Women underwent CS during a three-month period at each maternity unit, between 1. October 2021 and 31. August 2022. Exclusion criteria: unavailable records or unknown indication. Methods: Case files of CS were audited against pre-defined, localised criteria. Main Outcome Measures: CS rate, indications and proportion of non-medically indicated CSs. Results: The CS rate was 31% (2949/9364); 2674/2949 (91%) CSs were included for analysis. Main indications were previous CS (1133/2674; 42%), prolonged labour (746/2674; 28%), and fetal distress (554/2674; 21%). Overall, 1061/2674 (40%) did not comply with audit criteria, main reasons being one previous CS with no trial of labour (526/1061; 50%); reported prolonged labour without actual slow labour progress (243/1061; 23%); and fetal distress with normal FHR (120-160 BPM) at time of decision (225/1061; 21%). Conclusion: Two in five CSs were categorised as non-medically indicated at time of decision. Particularly, fear of poor outcomes and delay in accessing emergency surgery may cause resource-consuming ‘defensive decision-making’ for CS. Investments in conducive urban maternity units are crucial to ensure safe vaginal births and to reach a population-based approach to ensure best possible timely care for all with the limited resources available. Funding: Danida Fellowship Centre, Denmark (18-08-KU), Aarhus University Research Foundation and Laerdal Global Health (2021-0095; 40662).