The South Oaks Gambling Screen (SOGS) was originally developed to screen for pathological gambling in clinical settings; however, its use has expanded to other purposes, settings, and populations, including prevalence studies of pathological gambling in the general population. Questions have been raised about the accuracy of the SOGS with its use in these new settings and populations. The purpose of this study is to examine current estimates of the reliability, validity, and classification accuracy of the SOGS in two different samples: (a) a general population sample (N=803) and (b) a gambling treatment sample (N=1589). DSM-IV diagnostic criteria served as the standard against which to measure the SOGS classification accuracy and both the SOGS and DSM-IV diagnostic criteria were administered to participants. The SOGS was found to have satisfactory reliability with coefficient alphas of .69 and .86 in the general population and gambling treatment samples, respectively. The SOGS demonstrated satisfactory validity by differentiating between the general population and the gambling treatment sample and by exhibiting high correlations with DSM-IV diagnostic criteria and moderate correlations with other measures of gambling problem severity. The SOGS demonstrated good to excellent classification accuracy in the gambling treatment sample, but had poorer accuracy in the general population sample with a 50% false positive rate. The SOGS overestimated the number of pathological gamblers in the general population, as compared to DSM-IV diagnostic criteria.