AbstractBackground/purposeStage at diagnosis is an important metric in treatment and prognosis of cancer, and also in planning and evaluation of cancer control. In sub‐Saharan Africa (SSA), for the latter, the only data source is the population‐based cancer registry (PBCR). For childhood cancers, the 'Toronto Staging Guidelines' have been developed to facilitate abstraction of stage by cancer registry personnel. Although the feasibility of staging using this system has been shown, there is limited information on the accuracy of staging.MethodsA panel of case records of six common childhood cancers was established. A total of 51 cancer registrars from 20 SSA countries staged these records, using Tier 1 of the Toronto guidelines. The stage that they assigned was compared with that decided by two expert clinicians.ResultsThe registrars assigned the correct stage for 53%–83% of cases (71% overall), with the lowest values for acute lymphocytic leukaemia (ALL), retinoblastoma and non‐Hodgkin lymphoma (NHL), and the highest for osteosarcoma (81%) and Wilms tumour (83%). For ALL and NHL, many unstageable cases were mis‐staged, probably due to confusion over the rules for dealing with missing data; for the cases with adequate information, accuracy was 73%–75%. Some confusion was observed over the precise definition of three stage levels of retinoblastomas.ConclusionsA single training in staging resulted in an accuracy, for solid tumours, that was not much inferior to what has been observed in high‐income settings. Nevertheless, some lessons were learned on how to improve both the guidelines and the training course.