Problem gambling is a non-substance-based addictive disorder that can cause significant distress and dramatic consequences. Despite extensive research in neuroscience and clinical/social psychology, few contributions have been made from formal models of behavioural economics. We apply Cumulative Prospect Theory (CPT) to provide a formal analysis of cognitive distortions in problem gambling. In two experiments, participants made decisions between pairs of gambles and completed a standard gambling assessment. We estimated the parameter values specified by CPT for each participant and used those estimates to predict gambling severity. In Experiment 1, severe gambling behaviour was associated with a shallow valuation curve, a reversal of loss aversion, and decreased influence of subjective value on decisions (i.e., more noise or variability in preference). Experiment 2 replicated the effect of shallow valuation but did not demonstrate reversed loss version or noisier decisions. Neither experiment provided evidence of differences in probability weighting. We explore implications of the findings and conclude that problem gambling at least partially reflects a fundamental distortion to subjective valuation.