South Africa, like most developing countries, suffers from lack of complete vital statistics recording system. Nevertheless, regular collection of nationally represented data, with the inception of annual October Household Surveys in 1993 help to indirectly measure the country’s dynamic processes. New samples are drawn from these independent surveys so that it is possible to follow cohorts of individuals over time. It is of interest, therefore, to examine whether the various changes observed are real or result from differences in the sampling designs. This study uses the sample age-cohort means to identify marital behaviour of primeaged indigenous women in South Africa. Simple graphical techniques are used for this purpose. The study identifies a shift in marriage patterns between the younger and older cohorts, with a higher proportion of the older cohorts marrying than the younger cohorts. Further evidence from the decomposition analysis using age-period-cohort models indicate that period effect is constant around zero across the period under study while the age and cohort effects are non-zero and significant, suggesting that the observed shift in marital behaviour is real.