This paper presents a theory of firm training that explains why firms invest in general training, even though workers do not pay for it. The key idea is that the current employer has superior information about the worker's ability, giving it monopsony power over the worker. This encourages the firm to provide training. The model can lead to multiple equilibria: one with high training and low quits, and another with low training and high quits. Using microdata from Germany, the paper shows that the predictions of the specific human capital model are rejected, while the model receives support from the data. The paper also discusses the role of adverse selection and raids by other firms. It shows that raids can lead to an extreme form of the winner's curse, making this strategy unprofitable. The model predicts that military quitters earn more than voluntary quitters because they are freed from monopsony power. The paper also contrasts its findings with a standard human capital model, showing that they cannot easily be reconciled. The paper concludes that the model has important implications for understanding training and labor market dynamics. It suggests that the organization of the labor market plays a crucial role in determining training levels. In markets where poaching is restricted, training levels should be higher than in markets where raids are not restricted. The paper also highlights the importance of verifying training levels and the role of adverse selection in determining training and wage outcomes.This paper presents a theory of firm training that explains why firms invest in general training, even though workers do not pay for it. The key idea is that the current employer has superior information about the worker's ability, giving it monopsony power over the worker. This encourages the firm to provide training. The model can lead to multiple equilibria: one with high training and low quits, and another with low training and high quits. Using microdata from Germany, the paper shows that the predictions of the specific human capital model are rejected, while the model receives support from the data. The paper also discusses the role of adverse selection and raids by other firms. It shows that raids can lead to an extreme form of the winner's curse, making this strategy unprofitable. The model predicts that military quitters earn more than voluntary quitters because they are freed from monopsony power. The paper also contrasts its findings with a standard human capital model, showing that they cannot easily be reconciled. The paper concludes that the model has important implications for understanding training and labor market dynamics. It suggests that the organization of the labor market plays a crucial role in determining training levels. In markets where poaching is restricted, training levels should be higher than in markets where raids are not restricted. The paper also highlights the importance of verifying training levels and the role of adverse selection in determining training and wage outcomes.