In an evaluation of a job-training program, the influence of the program on the individual earnings capacity is important, because it reflects the program effect on human capital. Estimating these effects is complicated because earnings are observed for employed individuals only, and employment is itself an outcome of the program. Point identification of these effects can only be achieved by usually implausible assumptions. Therefore, weaker and more credible assumptions are suggested that bound various average and quantile effects. For these bounds, consistent, nonparametric estimators are proposed. In a reevaluation of Germany's training programs of 1993 and 1994, we find that the programs considerably improve the long-run earnings capacity of its participants.