Recent theoretical work on poverty traps shows that when there are increasing returns to investments in human capital, initial conditions strongly influence both the distribution of human capital and wealth (Galor and Zeira, 1993; Ljungqvist, 1993; Mookherjee and Ray, 2002). There is mounting evidence across both developed and developing nations suggesting a strong empirical basis for assuming convexity of the return structure of human capital investment.1 The available evidence for South Africa is consistent with this finding (see appendix A1).2 However,it is difficult to establish robustness of the result for several reasons. Firstly, much of the earlier work on the relationship between earnings and schooling relies on unrepresentative samples of data collected during Apartheid (most notably the various population censuses dating to the early 1970s). Secondly, even though a vast literature has appeared on the subject over the last decade, made possible in part through the availability of nationally representative datasets such as the World Bank LSMS of 1993, the methods employed vary tremendously, making it difficult to establish robustness.