Background The neurotrophic hypothesis postulates that mood disorders such as bipolar disorder (BD) are associated with a lower expression of brain-derived neurotrophic factor (BDNF). However, its role in peripheral blood as a biomarker of disease activity and of stage for BD, transcending pathophysiology, is still disputed. In the last few years an increasing number of clinical studies assessing BDNF in serum and plasma have been published. Therefore, it is now possible to analyse the association between BDNF levels and the severity of affective symptoms in BD as well as the effects of acute drug treatment of mood episodes on BDNF levels. Methods We conducted a systematic review and meta-analysis of all studies on serum and plasma BDNF levels in bipolar disorder. Results Through a series of meta-analyses including a total of 52 studies with 6,481 participants, we show that, compared to healthy controls, peripheral BDNF levels are reduced to the same extent in manic (Hedges' g = −0.57, P = 0.010) and depressive (Hedges' g = −0.93, P = 0.001) episodes, while BDNF levels are not significantly altered in euthymia. In meta-regression analyses, BDNF levels additionally negatively correlate with the severity of both manic and depressive symptoms. We found no evidence for a significant impact of illness duration on BDNF levels. In addition, in plasma, but not serum, peripheral BDNF levels increase after the successful treatment of an acute mania episode, but not of a depressive one. Conclusions In summary, our data suggest that peripheral BDNF levels, more clearly in plasma than in serum, is a potential biomarker of disease activity in BD, but not a biomarker of stage. We suggest that peripheral BDNF may, in future, be used as a part of a blood protein composite measure to assess disease activity in BD.