Methods used to estimate the prey consumption by large carnivores include direct continuous observation, stomach content analysis, carcass observations and scat analysis. Continual observations are widely considered the best approach to estimate large carnivore diets, with lions (Panthera leo) being no exception. Continual observation allows the recording of all prey encounters and biases inherent in the other approaches are minimised. However, continuous observations are not always feasible, and in situations where animals cannot be observed at all times, diets are often estimated from observed carcasses. This often leads to an over-estimation of large kills in the estimated diet. Alternative methods that are free of the constraints placed on continuous observations are needed to provide data of a similar quality to that obtained using these continuous observation bouts. I employed a cluster follow up technique to locate lion kills from remotely accessed Global Positioning System (GPS) data from lions in the Kruger National Park (KNP). I develop Generalized Linear Models (GLMs) that increase the probability of locating kills at GPS cluster events. By increasing the predictive ability of detecting kills I show that this technique can be used to locate kills in a more efficient manner than random searching of GPS clusters, with further advantages in that multiple groups of lions can be monitored simultaneously. By incorporating this technique into an adaptive research framework, the diet of lions (and that of other large carnivores) can be estimated. In addition, I show that the spatial association between lions at kill sites, while feeding on carcasses, provides a further increase in the predictive ability of kill site models. Lionesses were found to be considerably closer together at the start of clusters associated with kills in comparison to clusters where no kill was found. This pattern remained consistent for both small and large kills. This proximity approach could therefore be incorporated into the GLMs that are developed to predict kill sites of large social carnivores. To further reduce the bias (where small kills are often missed) inherent in carcass observations, I combined scats and carcasses collected from known times, locations and lion groups to construct a temporal kill record for each group of lions. By combining scats and carcasses I estimate that at least 50% of the small prey items, namely impala (Aepyceros melampus) and warthog (Phacochoerus africanus) were missed when GPS clusters were investigated for carcasses. Ultimately, I show that a combination of GPS cluster investigations based on models developed using GPS movement data in combination with lion proximity data, augmented with scats collected at GPS clusters, could provide estimates of large carnivore diets that begin to approach estimated diets obtained through continuous monitoring. The resulting diet, estimated from the GPS cluster approach in combination with scat collection, indicated that the dominant prey item in the region was zebra (Equus quagga) followed by wildebeest (Connochaetes taurinus), impala and buffalo (Syncerus caffer). Selection indices for the eight dominant prey items were calculated using prey availability measures obtained from the aerial census data and ground counts of groups. It has been suggested that group level selection is a better approach to calculating predator-prey interactions, and that stability in predator-prey systems is improved if group metrics of prey are used as apposed to individual measures of availability. I show that there is a considerable shift in selection indices, as well as in the order that prey is selected, when using different measures of prey availability. In selection studies, more effort needs to be paid to the assessment and definition of prey availability to ensure results accurately reflect selection patterns in the field, especially when data are used for the development of management practices. Combining buffalo predation data collected from GPS cluster investigations with buffalo mortality data collected over five years prior to the commencement of the GPS cluster investigations, allowed an investigation into patterns of lion predation on buffalo between 2000 and 2007. Buffalo of both sexes were more vulnerable to predation in habitats that gave lions an ambush advantage (i.e. increased grass height and tree density). Despite this similarity in landscape risk, different processes lead to similar fates in dangerous habitats for buffalo of both sexes. Predation pressure by lions on buffalo increased following periods of reduced rainfall; with more buffalo predated on following drier six month periods. Predation on males constituted a significant proportion of all predation and was focused predominantly into the late dry season. The resulting method of locating kills by using GPS clusters and correcting carcass data with scats collected along the movement path represents a robust technique to estimate large carnivore diets. In the concluding chapter I present avenues where future research can build on the current thesis and present a framework that can be employed when attempting to estimate large carnivore diets. / Thesis (PhD)--University of Pretoria, 2010. / Zoology and Entomology / unrestricted