Using the degradation distribution method to establish a reliability evaluation model for performance degradation data requires a number of strict assumptions, which are difficult to meet in practical applications. In order to avoid the shortcomings of the degradation distribution method, a data-driven reliability evaluation model of degradation data was proposed. Firstly, the non-parametric adaptive kernel density estimation was constructed to obtain the probability density function of the product degradation at each moment, at last the three-parameter Weibull distribution was used to fit the reliability at each time to obtain the reliability curve of the product performance. The K-S test proves that the model is stable and superior to the parameter method based on the case data, and it is not easy to be aggressive. The analysis method is convenient for practical engineering application.