Abstract: The Trivers-Willard hypothesis (TWH) predicts that a mother will treat a son or daughter differently depending on her ability to invest and the impact of her investment on offspring reproductive success. Although many studies have investigated the hypothesis, few have definitively supported or refuted it because of confounding factors or an inappropriate level of analysis. We studied maternal investment in sons and daughters in feral horses, Equus caballus, which meet the assumptions of the TWH with a minimum of confounding variables. Population level analyses revealed no differences in maternal behaviour towards sons and daughters. When we incorporated mare condition, we found that sons were more costly to mares in good condition, whereas daughters were more costly to mares in poor condition, although no differences in maternal behaviour were found. However, since the TWH makes predictions about individual mothers, we examined investment by mares who reared both a son and a daughter in different years of the study. Mares in good condition invested more in their sons in terms of maternal care patterns, costs to maternal body condition and costs to future reproduction. Conversely, mares in poor condition invested more in daughters. Therefore, with an appropriate level of analysis in a species in which confounding variables are minimal, the predictions of the Trivers-Willard hypothesis are supported. Copyright 2000 The Association for the Study of Animal Behaviour.