ABSTRACT:The specification of a model for selection to labour market training is studied. Based on present value maximization, it is demonstrated that frequently adopted specifications of wage equations yield selection models containing a difference between alternative wage rates paid in the training period. In addition, the training effect on future wage rates can be sign determined. Using this, we test the hypothesis of present value maximization. A test that is robust against omitted variables and heteroskedasticity is used. Based on a sample of unemployed Swedish workers, we find positive training effects for most sub‐group