The reliability of many highly reliable products is usually evaluated by two or more performance characteristics. To address this concern, a reliability analysis method based on Mahalanobis distance (MD) is proposed. Firstly, MD is used to transform the bivariate degradation data to one dimensional MD data. The MD threshold is obtained by the Monte-Carlo method Then, a simple wiener process model and three wiener process models with random effects are investigated. The statiscal inference is performed via Bayesian approach. An integrated method of model selection is given, which considers not only the hypothesis testing model and the quantile-quantile (QQ) plot, but also the refined leave-one-out cross-validation value. Finally, for an illustration of the proposed method, a numerical example about fatigue cracks is discussed and some results are presented.