Flavor is a major determinant of consumption. Although commonly referred to as “taste”, flavor is a multisensory experience: drawing from gustatory, olfactory and somatosensory inputs, each sourced from separate peripheral senses. Previous work in humans demonstrates that multisensory flavor cues are integrated to inform perceptual decisions. However, the computations underlying multisensory flavor interactions and their role in food choice remain poorly understood. Drawing from previous work on other multisensory systems, we evaluated the validity of the maximum likelihood estimation (MLE) framework in explaining multisensory flavor preference decisions. We used rats as a model system to obtain preferences in a series of daily two-alternative free choice tests, in which animals drank from two bottles containing taste, odor or taste+odor mixture solutions. Mean preference and variance over repeated presentations of the same condition (“reliability”) were analyzed. The MLE framework predicts that judgments of multisensory stimuli are a weighted average of the unisensory component judgement; that the weight components carry is proportional to their relative reliability; and that multisensory judgements are more reliable than their component judgements. Results from naïve rats (raised on standard chow) confirm these predictions. Results from rats raised with a limited set of specific taste-smell mixtures further show that the ability to weight taste and smell components is unaffected by flavor congruency. Finally, results from rats raised on a diet consisting of a wide variety of real foods show that sensory enrichment increases the overall weight animals place on odor components. This work provides a quantitative framework for understanding the multisensory interactions underlying hedonic evaluation of flavor and the factors that shape them, and suggest specific hypotheses regarding their neural underpinnings.