In order to solve the health evaluation problem of self-validating pneumatic actuators, a data-driven method was proposed. A prediction model was constructed using relevance vector machine (RVM) regression based on the normal working data of actuators. The residual feature was obtained as the event set by subtracting the actual output of the actuator from the predicted results. The target countermeasure set was established as consisting of four evaluation indexes of health, sub-health, marginal failure, and failure. The normal distribution function and the semi-trapezoidal function were selected as the membership functions to build the benchmark models that express the performance degradation degree of the actuator. The weight distribution models of local health degree and comprehensive health degree were established by using the analytic hierarchy process, grey relation algorithm, and entropy method. Finally, the least squares support vector machine was used to determine the health level. Results show that the method realized the overall and partial health assessment of pneumatic actuators, which is more practical and can reflect the performance status of self-validating pneumatic actuators.