The fatigue failure of thermoplastic ankle–foot orthoses (AFOs) was observed in clinics. However, there was no standard evaluation for the AFOs to enhance the understanding of how AFOs become more readily acceptable to patients. Therefore, this study aimed to develop an ankle–foot simulator (AFS) as a testing apparatus for AFOs, and performed a pilot test to investigate the failure mechanism of anterior ankle–foot orthosis (AAFO). The accuracy and repeatability of the AFS during cyclic walking, cyclic stepping and cyclic stepping with the AAFO in sagittal plane were measured. The root mean square errors (RMSEs) of cyclic walking of AFS compared to a target gait data were less than 80.52N and 2.55° in the vertical ground reaction force and in the kinematics, respectively. The RMSE of ankle plantarflexion and dorsiflexion of AFS in the cyclic stepping tests were less than 1.25°. The repeatability was assessed by standard deviation, which were less than 9.46N and 0.72° in all testing conditions. A typical failure progression of five AAFOs was observed and graded for four phases under cyclic stepping test. Failure always initiated at the junction of anterior tarsal bar and lateral (or medial) bar of the AAFOs, from which the rest failures were extended. It is suggested that this junction must be reinforced or prevented the stress concentration to elongate the endurance of AAFO.