We use data from the 1993 Project for Statistics on Living Standards and Development to estimate the union wage premium of black workers in South Africa. Unionised black males (females) earn 26 (81) per cent higher monthly wages than those who are not unionized. For hourly wages, the union premium increases to 65(127) per cent for males (females). Controlling for a worker’s characteristics and assuming exogenous union membership and no sample selection, the hourly wage union premium equals 38 (19) percent for males (females). We then estimate a bivariate probit to investigate labour market participation and union membership decisions, and use those results to account for samples selections when estimating a Mincerian wage equation. The wage gain of a representative non-unionised worker who joins a union equals 123 (96) per cent for males (females). Using bootstrapping we construct the confidence intervals for those estimates and conclude that male and female union wage premiums are not statistically different. However, our results indicate that the South African union wage premium have been underestimated in the literature. A union wage premium in the neighbourhood of 100 per cent confirms that the South African labour market is highly segmented.