This paper studies the factorial structure of the 12-item version of the General Health Questionnaire (GHQ-12), Dutch translation, in a large Belgian sample (n = 9442) that is representative of the general population. Data was collected from the 2001 Belgian Health Interview Survey. In total, I I factor-structure models are tested by means of confirmatory factor analysis. Two models, a two-factorial model without cross-loadings and a three-factorial model with cross-loadings, prove to have acceptable model-fit. Given parsimony and the uni-dimensionality of the subscales, the authors prefer the two-factorial model over the three-factorial one. In a next step, it is examined whether a general, second-order factor can be seen underlying both the two- and three-factorial models. This, however, was only true for the three-factorial model. Based on correlations with other scales (the revised Symptom Checklist 90, SCL-90-R; the MOS Social Support Survey, MOS-SSS) and a variable on social contacts, the authors suggest that the 'social dysfunction' sub-scale be renamed 'dysfunction'. Copyright (c) 2005 John Wiley & Sons, Ltd.