I. Problems and approaches in the analysis of the motivation (internal stimuli) of animal behaviour are reviewed and the nature and advantages of stimulus-response models in dealing with complex cases are discussed. 2. The data on the motivation of fish feeding are reviewed by considering the topics of homeostasis, deprivation and satiation, systemic need versus gastric volume, and preference and selectivity. 3. Based on these observations, a tentative model is proposed. In this model hunger is presumed to be determined jointly by the metabolic debt (systemic need) of the fish and the amount of food in the stomach (gastric volume), two parameters which are themselves affected by many physiological and ecological factors. For given environmental conditions hunger determines the probability that an encountered item is eaten and the interval between encounters. From these parameters monophagous encounter and feeding rates can be calculated. In the case of polyphagy the rates can be calculated provided the relative availabilities of the food types are known. 4. It is shown that the model clarifies several points in the literature and predicts certain response relations present, but apparently unrealized, in the published data. Further, the model makes a number of untested predictions. 5. Some of these predictions were tested and confirmed in a controlled experiment involving pumpkinseed sunfish (Lepomis gibbosus), in individual aquaria, fed in a standard Latin square design with added control conditions. The fish were fed three different proportions of housefly adults and pupae after one or two days of deprivation. Food items were presented one at a time over a 30 minute feeding bout. The changes in behaviour following changes in deprivation and relative availability of food types were as predicted by the model. The key prediction that response measures would be related in a manner invariant over changes in the independent variables was also supported despite nonstationarities in several response measures. 6. The modelling approach, traditional drive theory, and the case study of fish hunger were used to develop a rigorous general definition of "drive" based on stimulus-invariant response relations. The preferred nature of such a concept for dealing with bodies of motivational data containing such relations is shown. The consequent overall possibilities of modelling diverse motivational systems are considered.