The Problem Gambling Severity Index (PGSI) was designed to assess gambling problems in community samples. It has 12 items, nine of which are scored to determine problem gambling level. Although recent studies have evaluated the PGSI's psychometric properties, these have not used methods best suited for this task. The present study used both confirmatory factor analysis (CFA) and item response theory (IRT) modelling to assess these properties for a college student sample (N = 366). The scale was unidimensional and had varying item factor loadings that were identical for male and female participants. The latent gambling problem characteristic obtained from the 12-item set was more strongly correlated with gambling frequencies, was reliable and provided more information than a scale comprised of unweighted items or the latent characteristic obtained from, or a scale comprised of, the scored items. Future research should consider using the weighted 12-item scale to assess gambling problems.