Depression is a common comorbidity in diabetes but conventional antidepressant treatments do not consistently improve outcomes. We tested whether established diabetes treatments can also improve depressive symptoms and examined biological correlates of response. We performed a multi-database systematic search of all clinical trials, which measured the effect of licensed diabetes treatments on depressive symptoms using a validated questionnaire. Results of randomised controlled trials (RCT's) were pooled for meta-analysis. Data were also collected on insulin resistance (HOMA-IR), C-reactive protein (CRP) and fasting blood glucose (FBG) as correlates of response. Nineteen studies (n = 3369 patients) were included in the qualitative synthesis, 9 testing thiazolidenediones, 5 metformin, 2 thiazolidenediones against metformin, 2 incretin-based therapies and 1 insulin. Most studies were of good quality. In random-effects meta-analysis of RCT's, pioglitazone improved depressive symptoms compared to controls (pooled effect size = -0.68 (95% C.I. -1.12 to -0.24), p = .003, N = 8, I = 83.2%). Conversely, metformin was comparable to controls overall (pooled effect size = +0.32 (95% C.I. -0.23 to 0.88), p = .25, N = 6, I = 94.2%), although inferior to active controls (pooled effect size = +1.32 (95% C.I. 0.31-2.34), p < 0.001, N = 3, I = 90.1%). In random-effects meta-regression, female sex (β = -0.023, (95% C.I.-0.041 to -0.0041), p = .016, N = 8) predicted reduction in depressive symptoms with pioglitazone, but baseline HOMA-IR, FBG and severity of depressive symptoms did not. In conclusion, pioglitazone was associated with improvement in depressive symptoms, an effect more marked in women and poorly explained by effects on glycaemia and insulin resistance. Metformin had no consistent benefit on depressive symptoms. Further mechanistc trials of diabetes treatments as potential antidepressants are needed, stratified by sex and including serial measures of innate inflammation.