OBJECTIVES: To review the methodology of studies evaluating pregnancy/birth/child development outcomes of prenatal antidepressant exposure (PAE) and, when appropriate, assess the impact on study results using meta-regression. METHODS: A Medline search was conducted up to October 2011, restricting to English and human literature. The MeSH term “antidepressive agents,” was combined with MeSH terms for spontaneous abortion, preterm birth(PB), birth weight, small for gestational age, or poor child development. We added relevant articles from reference lists of reviews and included studies. We proposed that antidepressant class, type of data source, adjustment for maternal depression, exposure definition, duration and timing, and adjustment for more than three covariates would impact study results. Cochran’s Q statistic was used to test for heterogeneity of results. When heterogeneity existed, meta-regression using a random-effects model was performed to examine the impact of different methodological choices on study results. Meta-regression is a regression analysis in which each study is a unit of observation and the effect estimate is regressed on study characteristics. RESULTS: Of 33 studies included, we found differences in the definition of PAE, data sources, and covariates adjusted. Thirteen studies focused on selective serotonin reuptake inhibitors. Five studies defined exposure as one prescription of antidepressant during pregnancy versus 13 studies required PAE for at least one trimester. Pregnancy information services, birth registries, and referrals from hospitals/clinics were the most common data sources. Only 10 studies adjusted for maternal depression. Cochran’s Q was only significant for PB.  Meta-regression showed that adjusting for more than three covariates significantly (p=0.02) impacted pooled odds ratios (POR). POR for PB was 1.39 (95% confidence interval (CI)=1.27-1.52) for studies with 3 or more covariates versus 1.82(95% CI=1.43-2.29) for other studies. CONCLUSIONS: Methodological variations can make pooling of observational studies inappropriate. Consistency of study methodology should be assessed before combining studies even if heterogeneity were absent.