People often keep engaging in behaviors that used to be successful in the past but which are knowingly no longer effective in the current situation, so called action slips. Such action slips are often explained with stimulus-driven processes in which behavior is caused by a stimulus-response association and without information about the outcome of the behavior. This process is contrasted with a goal-directed process in which behavior is selected because it is expected to lead to a desired outcome. Failing to act in line with changes in the outcome, is taken as evidence for stimulus-driven processes. Stimulus-driven processes are assumed to get installed after overtraining and to be deployed under poor operating conditions. In line with this, previous research has found that action slips are more likely to occur after extensive training and when under time pressure. We propose an alternative goal-directed explanation according to which action slips are not caused by a stimulus-driven process, but rather by a goal-directed process that relies on old, no longer accurate, outcome information. In the current study, participants learned four stimulus-response-outcome contingencies during a single (i.e., moderate training) or a four-day training schedule (i.e., extensive training). Afterwards two contingencies were reversed. Results show that after extensive training and when under time pressure, participants not only committed more action slips but also reported more old response-outcome contingencies in line with these action slips. This is consistent with the goal-directed explanation that action slips result from a reliance on old, no longer accurate outcome information.