Three hundred and sixty three medium security federal prison inmates were interviewed with the South Oaks Gambling Screen (SOGS) with 5.2% of the sample achieving SOGS scores greater than 4 and another 7.4% attaining scores of 3 or 4, traditional markers of pathological gambling and problem gambling, respectively. Attempts were made to evaluate the relative accuracy of the continuum and dichotomy models of problem gambling by contrasting three groups of subjects (SOGS<3, SOGS=3 or 4, SOGS>4) on a series of gambling-related measures and comparing the relative ability of the SOGS and a dichotomized version of the SOGS to account for residual variance in these same gambling-related measures. Results were consistent with the view that problem gambling embodies features of both a continuum and dichotomy. Nearly half the subjects had desisted from problem gambling on their own, although self-remitters displayed less severe and extensive patterns of problem gambling than non-remitting subjects.