Objectives - a) To investigate the factor structure of the 12-item General Health Questionnaire (GHQ-12) in a sample of 18-year-old males; b) to test the ability of the standardized factor scores in discriminating between subjects who were suffering from emotional disturbance according to the psychiatrist's ratings and those who were not. Design - A sample of 363 subjects was selected during the medical examination for military service and invited to fill in the GHQ-12 and to be independently interviewed by a psychiatrist. Completed data were available for 320 subjects (88% of the selected sample). Results - The unrotated factor matrix allowed for the identification of two principal components with eigenvalues greater than 1, jointly accounting for 46.7% of the variance in the data. The first component, on which each item of the questionnaire showed a positive coefficient, was interpreted as measuring the overall severity of psychiatric disorder among respondents. The second component was bipolar with «positive» items having positive coefficients and «negative» items having negative coefficients. These findings suggest that the GHQ can assess both the negative (symptoms) and positive (subjective well-being) aspects of mental health. After Vari- max rotation each item of the questionnaire loaded significantly on one factor only, except for item 12 whose loading was lower than 0.500 on both factors. Factor A contained the «negative» items of the GHQ-12 and was termed as «general dysphoria»; factor B included 5 of the 6 «positive» items and was termed as «well-being/social functioning». The best discrimination between «cases» and «non-cases» was offered by the standardized factor scores on the «general Epidemiologia e Psichiatria Sociale, 2, 3, 1993 dysphoria» factor; no advantage was gained by the combination of the scores on the two factors. Conclusions - A better understanding of the GHQ-12 factor structure might provide useful information along with that offered by a single severity score resulting from the simple adding of positive responses and improve the description of emotional disorders beyond the traditional distinction between «cases» and «non-cases».