Heppner and Petersen (1982) derived a three-dimensional structure for the items of the Problem Solving Inventory using principal components analysis. D'Zurilla and Maydeu-Olivares (1995) pointed out the lack of any clear link of the scales of the Problem Solving Inventory (PSI) to social problem-solving theory, which impairs the interpretation of any empirical results obtained with the use of this inventory. To address this issue, we used maximum likelihood factor analysis to investigate the dimensional structure of this inventory. We found that an unrestricted three-dimensional model adequately fit this data but did not match Heppner and Petersen's structure, nor could it be interpreted substantively. Neither four- nor five-dimensional models could be interpreted regardless of the rotation procedure used. Although a two-factor model did not fit the data adequately either, a careful content analysis of the PSI items revealed only two interpretable clusters of items in this inventory, measuring problem-solving self-efficacy and problem solving skills, respectively. In addition, there is also a heterogeneous cluster of items responsible for the residual third dimension appearing in Heppner and Petersen's as well as in the current analyses. Using a theory-guided item-selection procedure we constructed two new PSI scales measuring the two interpretable constructs. The items composing these scales are shown to fit an independent clusters two-dimensional solution.