Background The stated intention to eliminate silicosis from the South African goldmining industry as well as current programmes to find and compensate ex-miners with silicosis require an understanding of variation in silicosis prevalence across the industry. We aimed to identify the predictors of radiological silicosis in a large sample of working miners across gold mines in South Africa. Methods Routine surveillance chest radiographs were collected from 15 goldmine "clusters" in a baseline survey undertaken in preparation for a separate tuberculosis isoniazid prophylaxis trial. All images were read for silicosis by a health professional experienced in using the International Labour Organisation (ILO) classification. Profusion thresholds of > 1/0 and > 1/1 were used. Demographic and occupational information was obtained by questionnaire. Predictors of silicosis were examined in a multivariable logistic regression model, including age, gender, racial ascription, country of origin, years since starting mine employment, mine shaft, skill category, underground work status and tuberculosis. Results The crude silicosis prevalence at ILO > 1/1 was 3.8% [95% confidence interval (CI) 3.5-4.1%]. The range across mine shafts was 0.8-6.9%. After adjustment for covariates, the interquartile range across shafts was reduced from 2.4 to 1.2%. Black miners [adjusted odds ratio (aOR) 2.8; 95% CI 1.1-7.2] and miners in full-time underground work (aOR 2.1; 95% CI 1.3-3.4) had substantially elevated odds of silicosis, while workers from Mozambique had lower odds (aOR 0.54; 95% CI 0.38-0.77). Silicosis odds rose sharply with both age and years since starting in the industry (p for linear trend < 0.005), with 95.5% of affected miners having > 15 years since first exposure and 2.2% < 10 years. Conclusions In surveillance of silicosis in working gold miners time since first exposure remains a powerful predictor. Age appears to be an independent predictor, while the detection of radiological silicosis in short-service miners requires attention. Public risk reporting by mines should include factors bearing on silicosis prevalence, specifically dust concentrations, with independent verification. Studies of silicosis and tuberculosis in ex-miners are needed, supported by an accessible electronic database of the relevant medical and dust exposure records of all gold miners.