We examined possible predictors of traumatic deaths by means of a case-control study nested within a large ongoing cohort study of residents of a Southern California retirement community. Baseline information was collected by means of a detailed mailed health survey completed by 11,888 residents in 1981-1982. In multivariate logistic regression analysis, being widowed or divorced (RR = 6.4, P = 0.009), sleeping 9 or more hours per night (RR = 4.6, P = 0.02), and drinking more than three alcoholic beverages a day (RR = 3.5, P = 0.04) were significant predictors of suicide risk. However, the strongest predictor of suicide was a mental outlook assessment summary score calculated from responses to seven questions derived from the Zung self-rating depression scale. Individuals in the poorest summary score category were 23 times more likely to commit suicide than individuals in the best summary score category (P = 0.004). Women who regularly practiced breast self-examination were also at very low risk for suicide (RR = 0.1, P = 0.0005). The mental outlook assessment summary score was also a predictor of accidental deaths, although the relative risk estimates were lower than those for suicide and the individual questions best predicting risk were different. A history of a serious chronic disease (RR = 2.6, P = 0.01) and moderate alcohol intake of less than two drinks per day (RR = 0.3, P = 0.01) were also predictors of accidental deaths. Estrogen use in women did not predict risk of traumatic deaths.