The difficulties of predicting overall meal acceptability when subjects have complete freedom to select from a wide range of possible foods are discussed, and some solutions offered in the context of a study conducted in a student cafeteria. We identified several subgroups in the data corresponding to different types of meals and modelled these separately. Tree diagrams were found to be a useful aid to identifying subgroups and provided a quick summary of the numbers of subjects selecting specific meal combinations. Some of the problems of using multiple linear regression to model overall acceptability are discussed, particularly the related problems of varied meal sizes and missing values inherent in a free selection environment.