Abstract: Against the backdrop of an accelerating aging society, social demands such as medical care and daily living assistance are continuously growing. As an important carrier of intelligent rehabilitation services, the application effectiveness of elderly rehabilitation robots directly affects the rehabilitation quality and safety experience of the elderly population. Based on the theoretical framework of human reliability, conducting a systematic analysis on the optimization of application effectiveness of elderly rehabilitation robots is of great practical significance for their safety management. By constructing a Bayesian network structural model and integrating multidisciplinary expert evaluation data, this study reveals the key constraining factors affecting system reliability and their mechanism of action. Using a function-oriented approach, a Bayesian network structural model is established on the Netica software platform, and quantitative analysis of influencing factors is realized through expert evaluation data. The sensitivity of each fault factor is evaluated based on the relative change rate between prior probability and posterior probability. Empirical research shows that the human reliability of elderly rehabilitation robots is mainly restricted by the following eight key factors: physical function limitations, excessive dependence or resistance, misjudgment of feedback signals, delayed or unclear feedback, noise interference, low personalization, insufficient family support, and inadequate social support. In response to the above issues, human reliability design strategies are proposed to improve the safety and practicality of the human reliability system of elderly rehabilitation robots.