Maternal investment (MI) was studied in Kaimanawa horses, a population of feral horses in central North Island, New Zealand. It is predicted that individual mares will vary their maternal behaviour so as to maximise their own and their offspring's reproductive success. Maternal investment is defined and measured by maternal input, proximal maternal costs and reproductive costs to the mother. The primary maternal input is milk. Time spent sucking is frequently used to measure milk intake based on the assumption that more sucking equates with proportionately more milk intake. A review found little support for this assumption and so I tested whether sucking time predicted milk intake by labelling the milk of thoroughbred mares with tritium and measuring its transfer to foals. No significant predictive relationship was found. Therefore sucking time cannot be used as an index of milk intake and conclusions about differential MI based on time spent sucking may be wrong. A mare's social environment is a significant modifier of her MI. Mares were more protective, and suffered reproductive costs, in bands with more than one stallion or in other circumstances where paternity is uncertain or rates of aggression are high. Despite individual differences in maternal style, mare behaviour was modified by experimental manipulations of the number of stallions in a band. In an unusual event when a mare and her adult daughter lived in the same band, both co-operated in the care of a single foal. Mare age and experience modified maternal behaviour. As mares age they become more successful at raising foals through better-targeted input, but no extra investment. I tested the Trivers-Willard hypothesis (TWH) of sex differential MI. The hypothesis predicts that mothers in better condition should produce more sons, and invest more in their sons, whereas mothers in poorer condition should favour their daughters. I argue that horses are an ideal species on which to test the TWH. We found no sex biased MI on a population level in terms of maternal input, proximal costs or ultimate costs. However, the TWH makes specific predictions about individuals, not populations. Individual mares in poor condition gave birth to more daughters and biased their MI towards daughters. Conversely, mothers in good condition gave birth to more sons and biased their investment towards sons, supporting all the predictions of the TWH. Mares alter their maternal investment in ways that conform to predictions based on maximisation of lifetime reproductive success.