South African higher education institutions are experiencing challenges regarding access, redress and the successful completion of programmes in an environment where there are still imbalances in the schooling system. Tools are needed that will assist with the process of selecting students. The aim of this study is to determine whether a test battery predicts academic success for postgraduate students at a historically disadvantaged university, and whether there are differences relating to gender and racial and language groups. The test battery considered, GRT2, was designed to measure three areas of ability--verbal, numerical and abstract reasoning. The sample consisted of an applicant group (774 candidates) and a student group (178 candidates). The internal consistency of the three subtests on the total applicant group was sufficiently reliable (above 0.8). Significant mean differences were found between the language groups and the race group for all three subtests, but only significant differences were found between the gender groups, with males obtaining higher scores. For the student group as a whole, correlations of all three subtests with both academic total and academic average marks were highly significant. In addition, the test exhibits predictive bias with regard to language, race and gender in the prediction of the academic total, and particularly for the numeric subtest. The results suggest that the test battery can help in identifying potentially successful students. However, cognisance must be taken of the differences between language groups and gender when interpreting test results. The paper contributes to the presently limited research on the validity of selection tools used in postgraduate programmes in African universities. (Contains 18 tables and 2 notes.)