Augmented reality (AR) overlays virtual information into a user’s real environment. AR software already exists for preoperative planning. User errors during planning can influence the surgery and put patients at risk. The aim of the study is to identify user errors of AR software for preoperative planning and to evaluate patient safety. A user test was conducted with 10 physicians who tested the AR software by planning a simulated surgery using the HoloLens 2 (Microsoft Corp., USA). The planning was divided into 21 tasks. An experimenter evaluated the execution of these tasks using a 3-step scheme (good, medium, poor). User errors that led to a medium or poor rating were examined in a Failure Mode and Effects Analysis. Each user error was assigned a risk classification: Unacceptable, conditionally acceptable, and acceptable. User acceptance was recorded using the System Usability Scale and interpreted using a benchmark scale. There were 64 user errors, corresponding to an error rate of 30%. Of these, 8 represented an unacceptable risk, and 16 a conditionally acceptable risk. User acceptance was interpreted as sufficient. User errors causing unacceptable risks occurred mainly in tasks requiring high precision. Therefore, preoperative planning should not be carried out completely in AR at present. AR software should currently only be used to visualize medical images.