This paper assesses the relationship between suicide and unemployment in Italy during the period 1977-1987, taking into account variations by gender and region. The first objective of the study is to provide descriptive longitudinal and cross-sectional aggregate-level analyses and also trends in individual-level and population risks for suicide in relation to unemployment. Our second objective is to use the Italian data to help discriminate between rival interpretations of the unemployment-suicide link, i.e. the operation of health selection or causal (susceptibility) mechanisms. Evidence for an association between suicide and unemployment among women was not convincing. The annual rate of female unemployment was negatively correlated with the female suicide rate and unrelated to the suicide rate among the unemployed, the relative risk or population attributable risk. Individual-level analyses for each year confirmed that unemployed women were more likely to commit suicide than their employed counterparts, although the overall relative risk was low (1.5) and conference intervals for six of the eleven annual risk ratios included unity (1). Among men, the unemployment rate was positively correlated over time with the suicide rate. However, change in the suicide rate across 18 geographic regions of Italy was unrelated to change in the unemployment rate, a finding which did not appear to be invalidated by a regression to the mean effect. Unemployment was also positively related to the suicide rate among the employed and population attributable risk, but unrelated to the rate among the unemployed or the relative risk. Comparison of suicide rates among the employed and unemployed revealed an excess of suicide among the latter in each year, with an overall relative risk of 3.4. On the basis of this contradictory and inconsistent evidence, we are cautious about offering definitive interpretations concerning the nature of the unemployment-suicide link among men. We conclude by suggesting the need for further individual-level studies using retrospective case-control methods.