This article examines the effect of the Massachusetts workforce development system on the earnings of disadvantaged adults using nonexperimental data from the late 1990s. The authors construct a comparison sample for program participants using individuals who apply for and are offered services yet do not participate in a training program. They present a series of difference-in-difference estimates that make several alternative efforts to correct for selectivity bias, including econometric models that regression adjust for observable characteristics and fixed-effect models that adjust for time-invariant person effects. They also employ probabilistic matching techniques to more finely align the treatment and comparison samples. On average, program participants experienced 20% increases in annual earnings 1 year postintervention and 25% increases after 2 years. The authors uncover considerable heterogeneity in these effects, suggesting that the most difficult to serve and the most job ready benefit the least.