This study aimed to examine sexual dimorphism in, and to produce a practical discriminant function for determining the sex of indigenous, Bantu-speaking, South African crania. The types of data to be used were a small number of traditional, or mathematically transformed three-dimensional, linear measurements, comparable to those in use by most physical and forensic anthropologists. The samples to be examined, separately and pooled, were of the Cape Nguni, Natal Nguni and Sotho subgroups. In addition, three local populations ('tribes'--Zulu, Xhosa and Southern Sotho) within these subgroups were also studied. Univariate male/female ratios indicate significant sexual dimorphism in the pooled South African crania. Canonical variates analysis of the pooled sample showed that facial width is the strongest discriminating morphometric variable; cranial length and basi-bregmatic height are the next most significant features. Eight measurements derived from the three-dimensional data were used to produce a series of discriminant functions for sex determination in the pooled sample, for which an accuracy of 77-80% was attained. Analysis of the calvaria and face, separately, has shown that the sex of damaged material can be diagnosed with a reasonable degree of accuracy (75-76%). The new functions for the pooled indigenous South African sample provide improved sex discrimination accuracy compared to those obtained by employing the commonly utilised statistics of Giles & Elliot (1963), even when a modified sectioning point is used. Functions calculated for the separate local populations gave variable and fairly low improvements in sexing accuracy. As the subdivisions at all levels are at present quite rapidly disappearing in South Africa, for most purposes it is now best to simply apply the pooled data functions for sexing crania.