In this paper, we econometrically examine the performance of salience theory (ST) for explaining observed behavior outside of a fully defined state contingent setting. Using a well-known dataset, we find that only a minority of people act consistently in the way proposed by ST when confronted with lottery choices for which only marginal probabilities are presented. By estimating the implied dependence structure of payoffs consistent with ST, only a minority of people infer independent payoffs when attaching probabilities to states, a finding at odds with ST. Instead, a majority treat lotteries as having positively correlated payoffs which raise questions about the independence assumption in ST. Finally, we also find that ST explains choice behavior less consistently than expected utility. Thus, ST should not be assumed to be superior to the most prominent models within the literature when employed outside of particular contexts.