BACKGROUND: Peripheral brain-derived neurotrophic factor (BDNF) is decreased in acute major depressive disorder (MDD) and bipolar disorder (BD) and recovered after treatment. Here we validated on a meta-analytical level whether BDNF restores differentially according to treatment response and whose measurements could be used as a biomarker, plasma or serum. METHODS: Using strict inclusion criteria, we compared BDNF in healthy controls and patients with MDD (38 studies, n=6619), and BD (17 studies, n=1447). Pre- and post-treatment BDNF levels were meta-analyzed according to treatment response in patients from 21 MDD studies (n=735) and 7 BD studies (n=88). Serum and plasma subgroups were analyzed, publication bias was assessed and heterogeneity was investigated. RESULTS: Serum and plasma BDNF were decreased in acute MDD and BD, and did not differ in euthymia in comparison with control subjects. Antidepressive treatment increased serum BDNF levels in MDD in responders (Cohen׳s d (d)=1.27, p=4.4E-07) and remitters (d=0.89, p=0.01), significantly more than in non-responders (d=0.11, p=0.69). For plasma BDNF in MDD and for BD, the evidence was insufficient for a meta-analysis. Although no significant difference was found between serum and plasma ES, variance of plasma ES was higher. LIMITATIONS: Between-study heterogeneity was explained only partially; signs of publication bias in serum studies. CONCLUSION: Serum BDNF might be regarded as a biomarker for the successful treatment of MDD. Serum measurements seem more reliable than plasma ones. Further research should focus on defining optimal time points for BDNF measurements and increase evidence for the usage of BDNF as a predictive biomarker in BD.