ABSTRACT People in situations of homelessness (PSH) have high levels of psychological impairment. Few studies have analyse the suitability of the instruments used to measure these levels. The aim of this study was to analyse the factorial structure and invariance (sex, age, nationality and housing status) of the 12-item General Health Questionnaire (GHQ-12) for a sample of PSH (n = 603) living in the Community of Madrid (Spain). A quantitative study was designed using a survey that include the GHQ-12 and other health measures (self-assessed health and diagnosis of mental illness). A comparison of three models (unidimensional, two correlated latent factors and bi-factor) showed that the bi-factor model provided the best goodness-of-fit indicators. A latent structure was identified in which each item was loaded into a general factor measuring a single GHQ-12 factor. The number of latent factors and the structure of factorial loads for GHQ-12 items were similar between groups. The GHQ-12 can be used as a screening instrument rather than as an outcome measure in different clinical dimensions. It is an appropriate instrument for assessing psychological impairment among PSH regardless of sex, age, nationality or housing status.