Abstract: To further elevate the reliability of braking system important components under dynamic conditions, it is necessary to improve the judgement accuracy rate of elementary event failure rate for rail transit vehicle braking system. Based on the multivariate regression model analysis, taking the actual failure data of a vehicle′s braking system, and considering the braking cylinder pressure and the law of train running speed changing with time, it is discovered that the relationship between braking cylinder pressure and time is generally in accordance with BidoseResp function, a braking system failure prediction model is thus established. After training with the actual failure data, the failure occurrence situation of key components is predicted using the failure prediction model. Prediction results show that the failure prediction model has high accuracy rate and applicability.