It is difficult to acquire the run-to-failure operating status information of device in the industrial process. How to use the censored data from this information to estimate operation reliability of device becomes vital. Accordingly, this paper proposes an estimation method of operation reliability that can carry out the estimation of instantaneous reliability and the prediction of remained useful life at the same time, by means of the censored data. The key technology is to build an estimation model of instantaneous reliability using the censored data. The three steps of modeling are as follows: Firstly, on-line signals of various types are measured during the operating process. Second, after degradation features are extracted from these signals, considering only censored data, Gaussian process for machine learning is employed to predict various degradation feature indicator values at the time of failure, and distribution hypotheses are adopted to obtain their probability density functions. Third, the Bayes