With the development of advanced technologies, the reliability evaluation of key components or equipments and preventive maintenance decision-making are becoming more and more important. Online reliability assessment and remaining life prediction using degradation data of real-time monitoring have been accounted as a hot trend in development. This article presents a means of conditional reliability evaluation by extracting characteristic parameters, which can represent the performance of the equipment in service. According to the time order, the physical performance measurements are regarded as time series. Exponential smoothing method is adopted to establish time series prediction model by computing exponential smoothing values. Performance parameters over a future time period are gained by the model constructed before. Experiments were carried out on a double row cylindrical roller bearing to get the vibration information and a mathematical model was built to forecast the future performance. The experiments proved the validity of the aforementioned method.