Human error in routine procedural tasks is often attributed to momentary failures to remember what step to perform. We argue that task-specific steps, which can be defined as actions required to achieve a particular goal across a variety of different devices, are far less prone to error than device-specific steps, which can be defined as actions that are required for the operation of the device but do not directly contribute to the goal. An experiment is reported that supports this distinction, showing that device-specific steps are more error prone than task-specific steps. Moreover, we argue that these errors reflect a failure of memory because the error rate for device-specific steps was sensitive to increased working memory load, while the error rate for task-specific steps was not. The current work demonstrates that a distinction between device- and task-specific steps can be effective in explaining error patterns observed on a specific task.