The lack of accurate population-based information on childhood cancer stage and survival in low-income countries is a barrier to improving childhood cancer outcomes. In this study, data from three population-based registries in sub-Saharan Africa (Abidjan, Harare and Kampala) were examined for children aged under 15. We assessed the feasibility of assigning stage at diagnosis according to Tier 1 of the Toronto Childhood Cancer Stage Guidelines for patients with non-Hodgkin lymphoma (including Burkitt lymphoma), retinoblastoma and Wilms tumour. Patients were actively followed-up, allowing calculation of three-year relative survival by cancer type and registry. Stage-specific observed survival was estimated. The cohort comprised 381 children, of whom half (n = 192, 50%) died from any cause within three years of diagnosis. Three-year relative survival varied by malignancy and location and ranged from 17% (95% Confidence Interval [CI] 6-33%) for Burkitt lymphoma in Harare to 57% (95% CI 31-76%) for retinoblastoma in Kampala. Stage was assigned for 83% of patients (n = 317 of 381), with over half having metastatic or advanced disease at diagnosis (n = 166, 52%). Stage was a strong predictor of survival for each malignancy; for example, three-year observed survival was 88% (95% CI 68%-96%) and 13% (4%-29%) for localised and advanced Burkitt lymphoma, respectively (P < 0.001). These are the first data on stage distribution and stage-specific survival for childhood cancers in Africa. They demonstrate the feasibility of the Toronto Stage Guidelines in a low-resource setting and highlight the value of population-based cancer registries in aiding our understanding of the poor outcomes experienced by this population. This article is protected by copyright. All rights reserved.This article is protected by copyright. All rights reserved.