Let be a preference relation on a convex set F. Necessary and sufficient conditions are given that guarantee the existence of a set {ul} of affine utility functions on F such that is represented by U (f) = ul (f) if f ∈ Fl; where each Fl is a convex subset of F. The interpretation is simple: facing a “non-homogeneous” set F of alternatives, a decision maker splits it into “homogeneous” subsets Fl, and acts as a standard expected utility maximizer on each of them. In particular, when F is a set of simple acts, each ul corresponds to a subjective expected utility with respect to a finitely additive probability Pl; while when F is a set of continuous acts, each probability Pl is countably additive. Keywords preference representation, subjective probability, nonexpected utility, integral representation, multiple priors, countable additivity