This paper develops a framework for analyzing unemployment in terms of variations in the nt.imber and distribution of people becoming unemployed and in individual probabilities of leaving unemployment. Contrary to the emphasis on exit probabilities in the recent macroeconomics literature, we present empirical evidence in support of the proposition that changes in the size and distribution of the inflow Into unemployment are the primary determinant of the unemployment rate. Instead of falling at the beginning of a recession, the outflow rate rises (with a lag) in response to the increased inflows which drive the recession. In contrast to normal unemployment, cyclical unemployment is concentrated in groups with low normal exit probabilities; so the observed procyclical variation in the average exit probability may largely he explained by predictable distributional effects.