A functional model of feeding behavior and physiology of suspension-feeding bivalves assumes that the net rate of energy gain from the available diet is maximized. Unlike previous models, a term is included for the rate of energy investment in digestive processes, such as the production of digestive enzymes and intracellular digestive products. In the model the digestive investment is applied within an absorption efficiency-gut residence time relationship. The choice of this relationship is based partly on two important assumptions. They are that the rate of digestion and absorption of the food should increase with the amount of digestive investment and that resorption of the digestive investment should depend on gut residence time. The choice of relationship has also been based on extensive empirical evidence of net enrichment of feces with endogenous metabolic products when gut residence times are short. The model suggests that the net rate of energy gain can be maximized through an optimal combination of gut residence time, digestive investment, and dietary selection coefficients. Analysis of the model with respect to variation in dietary quality and quantity suggests that feeding rate is constrained by the energetic costs of processing the food when food availability is low. When food is abundant, the rate of feeding is limited by constraints imposed by digestibility of the food and limitations on the rate of absorption of energy within the gut. The effect of a variable rate of digestive investment is to provide greater flexibility within the feeding system, particularly at the upper and lower extremes offood abundance. When particle concentrations and qualitities are such that behavioral dietary selection, coupled with increased rates of filtration, can be utilized to improve net energy gain, metabolic digestive investment is predicted to remain relatively constant. Relationships predicted by the model are in good agreement with empirical data. Here I develop an optimal feeding model for a suspension-feeding bivalve. Unlike previous models, a term is included which represents the amount of metabolic investment in the digestion process. The model is based on comprehensive data gained from the study of Mytiius edulis L., particularly on the digestive process (Hawkins et al. 1986), and evidence of metabolic fecal loss (Hawkins and Bayne 1985; Hawkins et al. 1990). Metabolic fecal loss is the net enrichment of the feces with metabolic products as revealed by short-term experiments on diets of poor quality. It is believed to reflect the investment in the digestive process (Bayne et al. 1988). Additionally, in Acknowledgments This research was supported by contracts EI-4140RX and MDE-4140RX from the Water Research Centre to the U.K. Natural Environment Research Council. The work would not have been possible but for the previous research, and the encouragement, of B. L. Bayne. John Widdows and John R. W. Harris also provided criticism of the model and manuscript. I am grateful to an anonymous referee whose comments helped to focus and clarify this contribution. light of evidence on the magnitude of costs of feeding in relation to ingestion rate (Widdows and Hawkins 1989), the role of costs in such models is considered. Models applicable to the feeding process of suspension-feeding bivalve molluscs have been of three types. First, there are empirical models fitted to experimental data on the feeding response to environmental variables (e.g. Crisp et al. 1985). These are of limited value, however, being purely descriptive. Second, Jorgensen et al. (1986, 1988,199O) have presented an elegant model of the water transport system of bivalves by treating it as a physical pump. The third approach is by application of optimality models. As homeostatic systems, organisms show physiological adaptation to environmental changes, such as those brought about by variations in dietary quality and quantity. Therefore an optimization approach is likely to be appropriate. Although originally introduced to microphagous aquatic feeding modes (Lehman 1976; Lam and Frost 1976), these models have been most fully developed in relation