In the wake of a large burden of silicosis and tuberculosis among ex-miners from the South African gold mining industry, several programmes have been engaged in examining and compensating those at risk of these diseases. Availability of a database from one such programme, the Q(h)ubeka Trust, provided an opportunity to examine the accuracy of length of service in predicting compensable silicosis, and the concordance between self-reported employment and that officially recorded. Compensable silicosis was determined by expert panels, with ILO profusion ≥1/0 as the threshold for compensability. Age, officially recorded and self-reported years of service, and years since first and last service of 3146 claimants for compensable silicosis were analysed. Self-reported and recorded service were moderately correlated (R = 0.66, 95% confidence interval 0.64−0.68), with a Bland−Altman plot showing no systematic bias. There was reasonably high agreement with 75% of the differences being less than two years. Logistic regression and receiver operating characteristic curve analysis were used to test prediction of compensable silicosis. There was little predictive difference between length of service on its own and a model adjusting for length of service, age, and years since last exposure. Predictive accuracy was moderate, with significant potential misclassification. Twenty percent of claimants with compensable silicosis had a length of service <10 years; in almost all these claims, the interval between last exposure and the claim was 10 years or more. In conclusion, self-reported service length in the absence of an official service record could be accepted in claims with compatible clinical findings. Length of service offers, at best, moderate predictive capability for silicosis. Relatively short service compensable silicosis, when combined with at least 10 years since last exposure, was not uncommon.