The General Health Questionnaire (GHQ-12) detects non severe psychiatric illnesses. Although commonly treated as a general index, the definition of its factorial structure is cause for debate. This study aimed to test this structure comparing three models: two frequently cited in the literature (one- and bi- factorial) and a third, also one-factor, which controls response bias due to the negative wording of items. A total of 1,180 people participated (300 undergraduates, 311 military policemen, 274 elementary school teachers and 295 members of the general population) answering the GHQ-12 and demographic questions. Confirmatory factor analyses showed that the one-factor structure, controlling by the wording effect, gathered the best fit indexes, except among the military. This structure showed greater than 0.80 reliability in all groups. It was concluded that the one-factor model of the GHQ-12 is more appropriate, however, future studies are required with people with different occupations and levels of mental health.