Regurgitated food sharing among vampire bats is a classic textbook example of reciprocity ("reciprocal altruism"). But many authors have contested both the notion that reciprocity explains vampire bat food-sharing and the importance of reciprocity more generally. In Chapter 1, I review the literature on evolutionary explanations of cooperation. I show why reciprocity was once considered important but is now considered rare: overly literal translations of game theory strategies have resulted in problems for both defining and testing reciprocity. In Chapter 2, I examine the relative roles of social predictors of food-sharing decisions by common vampire bats (Desmodus rotundus) under controlled conditions of mixed relatedness and equal familiarity by fasting 20 individuals in 48 trials over two years. The food-sharing network was consistent, symmetrical, and correlated with mutual allogrooming. Non- kin food-sharing patterns were not consistent with harassment or byproduct explanations. I next attempted to manipulate food-sharing decisions in two ways. In Chapter 3, I administered intranasal oxytocin to test for effects on allogrooming and food sharing. I observed that inhaled oxytocin slightly increased the magnitude of

food donations within dyads, and the amount of female allogrooming within and across all partners, without increasing number of partners. In Chapter 4, I assessed contingency of food-sharing in 7 female dyads (including four pairs of mother and adult daughters) with prior histories of sharing. To test for evidence of partner switching, I measured dyadic levels of food sharing before and after a treatment period where I prevented dyadic sharing (each bat could only be fed by others). A bat's sharing network size predicted how much food it received in the experiment. When primary donors were excluded, subjects did not compensate with donations from other partners. Yet, food-sharing bonds appeared unaffected by the non-sharing treatment. In particular, close maternal kin were clearly not enforcing cooperation using strict contingency. I argue that any contingencies within such bonds are likely to involve multiple services and long timescales, making them difficult to detect. Simple and dyadic `tit-for-tat' models are unlikely to predict cooperative decisions by vampire bats or other species with stable, mixed kinship, social bonds.