Abstract Objective To assess the quality of neonatal resuscitation (NR) through video recordings and identify potential areas for improvement. Design Prospective cohort study. Setting From September to December 2019, at four district hospitals in Pemba, Tanzania. Population All labouring women and their newborns were eligible for participation. Main Outcome Measures Videos were analysed for quality-of-care indicators based on the NR algorithm. Questionnaires on quality-of-care indicators were answered by health workers (HW) and mothers. Risk factors for neonatal mortality were analysed in a binomial logistic regression model. Results A total of 1440 newborns were enrolled. Within the neonatal period, 34 newborns died (23.6 per 1000 live births). During the study period, 90 neonatal resuscitations were performed, of which 20 were recorded. Meantime to initiate positive pressure ventilation (PPV) with bag-and-mask was 98 seconds (10 – 416 s), it was inadequately performed in 15 cases (75%). Half (10/20) did not have PPV initiated within the first minute, and in 1 case (5.0%) no PPV was performed. PPV was initiated but not sustained in 16/20 (80%) newborns. Of the 20 infants with videos, 10 died: Eight after failed resuscitation and two in the first 24 hours. The majority of HW 49/56 (87.5%) had received training in NR. Conclusions Despite nearly 90% of HW being trained in NR, video analysis revealed significant deviations from guidelines, provided direct evidence of gaps in the quality of care and identified areas for future education, particularly effective bag- and mask ventilation. Keywords PERINATAL NEONATAL MORTALITY RESUSCITATION VIDEO RECORDIN