The concept of device- vs. task-orientation allows to identify subtasks that are especially prone to errors. Device-oriented tasks occur whenever a user interface requires additional steps that do not directly contribute to the users’ goals. They comprise, but are not limited to, initialization errors and postcompletion errors (e.g. removing a bank card after having received money). The vulnerability of device-oriented tasks is often counteracted by making them obligatory (e.g. by not handing out the money before the bank card has been removed), making it even harder to predict where users will have problems with a given interface without dedicated user tests. In this paper we show how cognitive modeling can be used to predict error rates of device-oriented and task-oriented subtasks with respect to a given application logic. The process is facilitated by exploiting user interface meta information from model-based user interface development.