Maintenance has been one of the most important domains for augmented reality (AR) since its inception. AR applications enable technicians to receive visual and audio computer-generated aids while performing different activities, such as assembling, repairing, or maintenance procedures. These procedures are usually organized as a sequence of steps, each one involving an elementary action to be performed by the user. However, since it is not possible to automatically validate the users actions, they might incorrectly execute or miss some steps. Thus, a relevant open problem is to provide users with some sort of automated verification tool. This paper presents a system, used to support maintenance procedures through AR, which tries to address the validation problem. The novel technology consists of a computer vision algorithm able to evaluate, at each step of a maintenance procedure, if the user correctly completed the assigned task or not. The validation occurs by comparing an image of the final status of the machinery, after the user has performed the task, and a virtual 3D representation of the expected final status. Moreover, in order to avoid false positives, the system can identify both motions in the scene and changes in the camera’s zoom and/or position, thus enhancing the robustness of the validation phase. Tests demonstrate that the proposed system can effectively help the user in detecting and avoiding errors during the maintenance process.