Studies examining the effectiveness of welfare-to-work programs present findings that are mixed and sometimes at odds, in part due to research design, data, and methodological limitations of the studies. We aim to substantially improve on past approaches to estimate program effectiveness by using administrative data on welfare recipients in Missouri and North Carolina to obtain separate estimates of the effects of participating in sub-programs of each state's welfare-to-work program. Using data on all women who entered welfare between the second quarter of 1997 and fourth quarter of 1999 in these states, we follow recipients for sixteen quarters and model their quarterly earnings as a function of demographic characteristics, prior welfare and work experience, the specific types of welfare-to-work programs in which they participate, and time since participation. We focus primarily on three types of subprograms–assessment, job readiness and job search assistance, and more intensive programs designed to augment human capital skills–and use a variety of methods that allow us to compare how common assumptions influence results. In general, we find that the impacts of program participation are negative in the quarters immediately following participation but improve over time, in most cases turning positive in the second year after participation. The results also show that more intensive training is associated with greater initial earnings losses but also greater earnings gains in the long run.