The publication by Servaas van der Berg and his colleagues at the University of Stellenbosch last year of a set of estimates of poverty headcounts and gaps, pointed to substantial success in the battle against poverty in South Africa in the period 2000–2004, an improvement attributable mainly to massive increases in social grants. The present paper takes issue with the van der Berg et al headcount of 15.4 million in 2004, arguing that a more likely figure is in the region of 18–20 million. This difference is far too large to be ignored by policymakers. Using household survey data from Statistics South Africa (as opposed to the All Media and Products Survey data used by van der Berg et al), the paper develops a method for dealing with under-reporting by specifying limits to the size of the corrections that can be applied to income estimates without straying beyond the bounds of the plausible. The paper questions the uncritical assignment of all disability grant payouts to poverty alleviation. Although it concentrates on estimating the headcount and poverty gap in 2004, the paper does offer a tentative comparison of its results with those made by other researchers for the year 2000. The paper ends with recommendations for a series of changes that need to be made to the General Household and\nLabour Force Surveys conducted by Statistics South Africa, if these surveys are to fulfil their potential role as suppliers of information about poverty