The 2-year outcome of 178 patients attending a community-based mental health service was assessed from a multidimensional perspective. The study investigated: (1) the effect of disease-related characteristics (such as diagnosis and illness duration) and of a series of outcome variables measured at baseline (global functioning, psychopathology, social disability, quality of life and satisfaction with services) on total costs of care over 2 years; and (2) the effect of costs of care and outcome variables measured at baseline on the corresponding outcome variables at 2 years. To gain insight into the multivariate longitudinal dependencies among variables, we used graphical Gaussian chain models, a new multivariate method that analyses the relationship between continuous variables taking into account the effect of antecedent and intervening variables, to reveal not only direct but also indirect correlations. Outcome variables showed the tendency to segregate, both at baseline and follow-up, into two distinct groups: a clinician-rated dimension (given by global functioning, social disability and psychopathology) and a patient-rated dimension (given by service satisfaction and subjective quality of life). Higher costs at 2 years were predicted by higher psychopathology at baseline, diagnosis of psychosis and longer duration of illness. Baseline values for each variable were the main predictors of the corresponding values at two years. Improvement in satisfaction with life at follow-up was experienced in those subjects with a lower functioning at baseline. This study throws some light on the complex relationships between clinical, social and economic variables affecting the medium-term outcome of mental health care.