One of the difficulties in analyzing accelerated life testingdata is the model-based failure probability prediction.Choosing an inferior model yields inaccurate predictionsthat can be exaggerated by extrapolation. Furthermore,testing data are often naturally clustered in groups, thussome modeling exibility must be granted to handle both theintra-cluster and inter-cluster variations. To address theseproblems, we discuss a data fitting strategy in this paper bydeveloping a semiparametric model with random effectsand the Bayesian piecewise exponential inference method.The proportional hazard model and Weibull acceleratedfailure time model are examined and compared. Our resultsuggests that the Bayesian piecewise exponential model withrandom effects outperforms other models.