We evaluate the Mexican training program for the unemployed PROBECAT/SICAT using methods that control for observable and non-observable factors. Comparing the different results over time allows us to gauge the size and evolution of hidden bias. We also compute the average treatment effect and the treatment effect on the treated. Our approach reveals the evolution of the program’s selection mechanism and judges the consequences of its expansions and contractions. We find that the program has a small though significant effect on employability, but no effect on wages. The hidden bias is large but declines over time and the selection mechanism turns from negative to neutral. These two aspects seem to be related to an important structural change in the design of the program that took place during the period under evaluation. All these results lead us to conclude that a parametric method controlling for un-observables provides the most complete tool for evaluating this program.