Few long-term studies on bipolar disorder (BD) have investigated the incidence and risk factors of suicide attempts (SAs) specifically related to illness phases. We examined the incidence of SAs during different phases of BD in a long-term prospective cohort of bipolar I (BD-I) and bipolar II (BD-II) patients, and risk factors specifically for SAs during major depressive episodes (MDEs). In the Jorvi Bipolar Study (JoBS), 191 BD-I and BD-II patients were followed using life-chart methodology. Prospective information on SAs of 177 patients (92.7%) during different illness phases was available up to 5years. The incidence of SAs and their predictors were investigated using logistic and Poisson regression models. Analyses of risk factors for SAs occurring during MDEs were conducted using two-level random-intercept logistic regression models. During the 5 years of follow-up, 90 SAs per 718 patient-years occurred. The incidence was highest, over 120-fold higher than in euthymia, during mixed states (765/1000 person-years; 95% confidence interval [CI] 461-1269 person-years), and also very high in MDEs, almost 60-fold higher than in euthymia (354/1000 person-years; 95% CI 277-451 person-years). For risk of SAs during MDEs, the duration of MDEs, severity of depression, and cluster C personality disorders were significant predictors. We confirmed in this long-term study that the highest incidences of SAs occur in mixed and major depressive illness phases. The variations in incidence rates between euthymia and illness phases were remarkably large, suggesting that the question "when" rather than "who" may be more relevant for suicide risk in BD. However, risk during MDEs is likely also influenced by personality factors.