This paper investigates the testable restrictions on the time-series behaviour of consumption and asset returns implied by the consumption/portfolio choice problem of an infinitely-lived, representative agent. Intertemporal preferences are characterized by utility functions that generalize conventional, time-additive, expected utility. These generalizations of expected utility, allow for a clear separation of observable behaviour attributable to risk aversion and to intertemporal substitution, and also provide simple nested-tests of the expected utility hypothesis. Using monthly New York Stock Exchange returns data and consumption measured with either per capita expenditures on nondurables or nondurables and services, the expected utility model is rejected. The over-identifying restrictions implied by the non-expected utility model are tested and do not, in general, lead to rejections of the theory.