Do government provided training programs benefit the participants and the society? We address this question in the context of female immigrants who first learn the new language and then choose between working or attending government provided training. Although theoretically training may have several outcomes, most evaluations have focused on only one outcome of training: the expected wage. However, training might have no direct effect on wage, but, nevertheless, affect employment probability in higher paid jobs. In order to measure the return to government provided training, and overcome the above reservations, we formulate an estimable stochastic dynamic discrete choice model of training and employment. Our estimates imply that training has no significant impact on the mean offered wage in bluecollar occupation, but training increases the mean offered wage in white-collar occupation by 19 percent. Training also substantially increases the job offer rates in both occupations. Furthermore, counterfactual policy simulations show that free access to training programs relative to no training could cause an annual earnings growth of 31.3 percent. This large social gain (ignoring the cost of the program) comes mainly from the impact of training on the job offer probabilities and, consequently, on unemployment, and not, as conventionally thought, from the impact of training on potential earnings. Moreover, free access to training increases the average ex-ante expected present value of utility for a female immigrant at arrival (individual benefit) by 50 percent relative to the existing training opportunity.