Background: Despite impressive advancements in early interventions in psychosis, there is an urgent need of robust neurobiological markers to improve the predictive value of psychosis transition. Available structural imaging literature in the field is undermined by several methodological caveats and a number of confounders such as exposure to antipsychotic treatment. Methods: Fourteen voxel-based morphometry studies of antipsychotic-naive subjects at enhanced clinical risk for psychosis (high risk [HR]) or experiencing a first-episode psychosis (FEP) were included. Formal meta-analysis of effect sizes and “signed differential mapping” voxel-based meta-analysis were combined to control the results for sample sizes, strength of individual findings, and confounding variables. Results: Formal effect size meta-analysis indicated consistent gray matter (GM) reductions both in subjects at enhanced clinical risk for psychosis and in first-episode subjects when compared with control groups. Voxel-based meta-analysis showed GM reductions in the temporal, limbic prefrontal cortex within the HR group and in the temporal insular cortex and cerebellum within the FEP group. Psychosis onset was characterized by GM decreases in temporal, anterior cingulate, cerebellar, and insular regions. GM alterations in the temporal regions directly related to severity of psychotic symptoms. There was no publication bias. Heterogeneity across studies was low. Sensitivity analyses confirmed robustness of the above results. Conclusions: Vulnerability to psychosis is associated with consistent GM decreases in prefrontal and temporolimbic areas. The onset of full disease is accompanied by temporoinsular, anterior cingulate, and cerebellar GM reductions. Neuroanatomical alterations in temporal regions may underlie the clinical onset of psychotic symptoms.