Econometric evaluations of public-sponsored training programmes generally find little evidence of an impact of such policies on transition rates out of unemployment. We perform the first evaluation of training effects for the unemployed adults in France, exploiting a unique longitudinal dataset from the unemployment insurance system. Using the so-called timing-of-events methodology to control for both observed and unobserved heterogeneity, we find that training does not accelerate the exit from unemployment, but has a significant and positive effect on the duration of the subsequent employment spell. Accounting for training duration, we find that longer training spells cause longer unemployment spells, but also longer employment spells, suggesting that training improves the matching process between jobseekers and firms.