We use a randomized experiment to evaluate a large-scale active labor market policy: Turkey's vocational training programs for the unemployed. A detailed follow-up survey of a large sample with low attrition enables precise estimation of treatment impacts and their heterogeneity. The average impact of training on employment is positive, but close to zero and statistically insignificant, which is much lower than either program officials or applicants expected. Over the first year after training we do find training to have had statistically significant effects on the quality of employment, and that the positive impacts are stronger when training is offered by private providers. However, longer-term administrative data shows that after three years these effects have also dissipated.