Despite the n-back task’s apparent effectiveness as a working memory (WM) training task, its status as a WM assessment is questionable. We analyzed the accuracy and reaction time data of participants performing of an adaptive n-back training task and developed a computational model to describe this performance. Application of our model to n-back training data suggests that performance is consistent with a two-stage, familiarity and recollection account. Furthermore, our results suggest that interference resolution is an important determining factor for task accuracy, especially when responding to targets.