This study uses the principal-agent framework to investigate the trade-off between the benefits of internal control stemming from a reduction of the losses from inappropriate employee actions and the cost of implementing it brought about by the possibility of collusion that it creates. It is shown that, when the agents find it relatively easy to collude, implementing internal control reduces agency welfare, defined as the sum of expected payments accruing to the principal and the agents, even as, with positive transaction costs of collusion, it improves productive efficiency, defined as the expected output. As a result, the principal, under certain conditions, finds it in her best interest to use internal control as a threat instead of implementing it. When this is the case, the principal sometimes prefers to decrease the accuracy of the accounting information system. The analysis of the agents' side contracting indicates that, even if the principal can prevent explicit collusion, for some values of parameters the possibility of tacit collusion still results in a loss. The study also investigates the effect of the choice of organizational form on the value of internal control. The analysis of two different versions of the model demonstrates that, for a wide range of parameters, creating a hierarchical structure reduces, albeit does not eliminate, the loss from collusion -- i.e., internal control and hierarchical delegation are complementary instruments of organizational design. It is also shown that, when one agent is ex ante more likely to be efficient than the other, in most cases the principal optimally appoints to the supervisory position the one who is less likely to be efficient. As a result, the supervisor, in expectation, exerts a lower effort level than the subordinate and collects higher salary. / text