Because cerebral morphological abnormalities in major depressive disorder (MDD) may be modulated by antidepressant treatment, inclusion of medicated patients may have biased previous meta-analyses of voxel-based morphometry (VBM) studies. A meta-analysis of VBM studies on medication-free MDD patients should be able to distinguish the morphological features of the disease itself from those of treatment. A systematic search was conducted for the relevant studies. Effect-size signed differential mapping was applied to analyse the grey matter differences between all medication-free MDD patients and healthy controls. Meta-regression was used to explore the effects of demographics and clinical characteristics. A total of 14 datasets comprising 400 medication-free MDD patients and 424 healthy controls met the inclusion criteria. The pooled meta-analysis and subgroup meta-analyses showed robustly reduced grey matter in prefrontal and limbic regions in MDD. Increased right thalamus volume was only seen in first-episode medication-naive patients, and increased grey matter in the bilateral anterior cingulate cortex only in medication wash-out patients. In meta-regression analyses the percentage of female patients in each study was negatively correlated with reduced grey matter in the right hippocampus. By excluding interference from medication effects, the present study identified grey matter reduction in the prefrontal-limbic network in MDD. The subgroup meta-analysis results suggest that an increased right thalamus volume might be a trait directly related to MDD, while an increased anterior cingulate cortex volume might be an effect of medication. The meta-regression results perhaps reveal the structural underpinning of the sex differences in epidemiological and clinical aspects of MDD.