This paper analyzes the link between training quality and labor-market outcomes. Multiple proxies for training quality are identified from bidding processes in which public and private training institutions compete for limited public funding in Peru. Information about exact dates of program enrollment is analyzed to show whether the first-come-first-served assignment rule randomized eligible individuals across courses of varying quality. Generalized propensity score (GPS) is implemented to estimate dose–response functions in the context of multiple treatments. We find that beneficiaries attending high-quality training courses show higher earnings and better job-quality characteristics than either beneficiaries attending low-quality courses or nonparticipants. The returns are particularly robust for women, making the provision of high-quality training services cost-effective. Furthermore, the most important training attribute is expenditures per trainee. Class size and infrastructure are weakly related to the expected impacts, while teacher experience, curricular activities, and market knowledge seem to bear no relationship with the expected impacts. External validity was assessed by using five cohorts of individuals over an eight-year period.