In the information fusion system, information provided by sensors is always not fully reliable. Thus it is necessary to evaluate the reliability of sensors for processing information conveyed by them. Therefore, the posterior probability is defined based on the confusion matrix of sensors. The reliability of sensors is evaluted by analyzing the relation between the acquired evidence and posterior probability. The evidence provided by sensors is discounted by using reliability factors of sensors. The discounted evidences are combined by using the Dempster’s combination rule. Numerical simulation demonstrates that the performance of sensors can be evaluated better and the fusion result can be improved by reducing the influence of low-reliability sensors.