This paper estimates the relative efficiency of eight Swedish labor market programs in reducing the unemployment duration for participants. The analysis uses a hazard regression model and a uniquely large and rich administrative data set that contains all adult workers who became unemployed during 1995-1997. We find that programs in which the participants obtain subsidized work experience and training provided by firms, have better outcome than classroom vocational training. The relative efficiency is similar across demographic and skill groups and independent of the timing of the placement. A careful examination of the assignment process to programs reveals no self-selection, but substantial administrative-selection. However, the administrative-selection appears to be unrelated to the outcome and, thus, should not bias the results.