We explore the relationship between exposure to violence during childhood perpetrated by adults inside the home and educational outcomes in the context of higher than average rates of violence in Cape Town, South Africa and the disproportionate exposure to violence of young South Africans (black and coloured youth in particular). We match official police murder statistics at the neighbourhood level to the Cape Area Panel Study to provide a unique descriptive analysis of violence in Cape Town and we determine the extent of selection bias using matching techniques. Using three measures of educational outcomes (numeracy and literacy test scores, dropout and high school exam results), we: (i) estimate kernel density functions of continuous educational outcomes measures by race and exposure to violence during childhood; (ii) remove constant differences in unobserved family and neighbourhood background that may bias the results by using sibling and neighbourhood fixed effect models; (iii) check the robustness of our sibling fixed effect regressions by including birth order effects. In the neighbourhood fixed effect regressions, the measures of exposure to violence are significant and have a large negative effect on educational outcomes (with the exception of literacy scores). In the sibling fixed effect regressions, the effect remains for two of the four measures of exposure to violence during childhood. The measure of exposure to emotional violence during childhood is least affected by selection bias and the only measure robust to the inclusion of birth order effects.