Obesity has risen to epidemic levels in the United States and around the world. Global indices of obesity such as the body mass index (BMI) have been known to be inaccurate predictors of risk of diabetes, and it is commonly recognized that the distribution of fat in the body is a key measure. In this work, we describe the early development of image analysis methods to quantify regional body fat distribution in groups of both male and female wildtype mice using magnetic resonance images. In particular, we present a new formulation which extends the expectation-maximization formalism commonly applied in brain segmentation to multi-exponential data and applies it to the problem of regional whole body fat quantification. Previous segmentation approaches for multispectral data typically perform the classification on fitted parameters, such as the density and the relaxation times. In contrast, our method directly computes a likelihood term from the raw data and hence explicitly accounts for errors in the fitting process, while still using the fitted parameters to model the variation in the appearance of each tissue class. Early validation results, using magnetic resonance spectroscopic imaging as a gold standard, are encouraging. We also present results demonstrating differences in fat distribution between male and female mice.