Abstract Background The role of environmental risk factors in the development of bipolar disorder (BD) is not well characterized. We evaluate the prevalence, duration, and predictive value of environmental exposures for BD in longitudinal studies. Methods We conducted a systematic search of PubMed, Scopus and PsychINFO databases until April 01, 2015, using the following words in combination: prenatal exposure; maternal exposure; trauma; childhood abuse; alcoholism; cannabis; smoking; cocaine; central stimulants; opioids; uv light; pollution; global warming; vitamin d AND bipolar disorder. Additional references were obtained through cross-referencing. We included (1) longitudinal cohort studies or case-control studies nested within longitudinal designs; (2) studies of subjects without lifetime BD diagnoses at initial assessment and a diagnosis of BD at follow-up by clinical or structured assessment. Familial-risk studies were excluded. We tabulated details of study-design, exposure, diagnostic criteria, risk of bipolar disorder expressed as odd ratio (OR), relative risk (RR) or hazard ratio (HR). Results Of 2119 studies found, 22 met inclusion criteria. Risk factors identified can be grouped in 3 clusters: neurodevelopment (maternal influenza during pregnancy; indicators of fetal development), substances (cannabis, cocaine, other drugs – opioids, tranquilizers, stimulants, sedatives), physical/psychological stress (parental loss, adversities, abuses, brain injury). Limitations Heterogeneity of designs and methodology prevented the use of meta-analysis of the findings; studies did not provide sensitivity, specificity and predictive value of the risk factors identified; case-control studies classify cases based on diagnostic membership, but do not control for familial or genetic liability; methods for determining the exposures varied among studies. Conclusion Only preliminary evidence exists that exposure to viral infection, substances or trauma increase the likelihood of BD. Given the limited data available, the specificity, sensitivity and predictive value could not be computed. As exposures are sometimes amenable to prevention, further research is needed.