This paper examines household income inequality in the South African October Household Survey datasets between 1995 and 1999, the Labour Force Survey 2000, and the Income and Expenditure Surveys 1995 and 2000. The paper reflects both on changing patterns of income inequality in South Africa, and on the quality and comparability of the data employed. We employ several measures of income inequality, employing nominal income and expenditure data from South Africa over the 1995-2000 period. Results prove sensitive to the choice of welfare measure. Furthermore, results from income data and expenditure data provide contrasting results. On self-reported income data, our findings are that inequality measures increased over the 1995 - 2000 time period. While we do not attach much credence to the evidence for reasons attaching to data quality, there is nevertheless evidence suggesting that the general increase in inequality for the African race group also hides a decrease in inequality for the bottom 1/3 of the income distribution, and (more unambiguously) a widening of inequality for the middle 1/3 of the income distribution for Africans. There is also some evidence of a narrowing inequality amongst rich households for the population as a whole. This suggests that there is at least some evidence consistent with a successful redistribution of income from richest to poorest households, though this has not yet reversed the high aggregate level of inequality in South Africa. Evidence from inequality measures based on expenditure data reverse the findings based on self-reported income. Where there is evidence of an increase in inequality, in most instances this proves to be statistically insignificant. On some measures African as well as total population inequality has declined significantly, and for a number of racial groups inequality has remained unchanged. The central conclusion of the paper is therefore that there is much contradictory evidence that emerges from household level data on income inequality - suggesting that the choice of data set is non-trivial in drawing inference on income inequality in South Africa.\n \n