Abstract: In order to deeply ultimate the information imbedded in information system audit evidences and achieves pre-maintenance，a quantitative model of audit evidences for information system operation reliability prediction is developed.According to the Dempster-Shafer theory and framework of information systems audit，qualitative and quantitative audit evidences with uncertain information can be transformed into a uniform format，and then a reliability prediction model is built to forecast the reliability of the system in the future though a series of evaluating value.Finally，a generic optimal learning algorithm is provided to determine the parameters of the proposed model，and utility theory is used to further explain model’s prediction results.