This study emphasis the Bayesian analysis for survival data using Frechet distribution based on doubly type-II censored samples, when some of the sample values might have been contaminated at either or both extremes, because in a clinical disease the maturation time of disease cannot predict exactly. The Bayesian estimators based on reference prior are prosed and performances of these estimators are compared with maximum likelihood estimators via simulation study based on standard deviation by taking different sample sizes and different parametric values. A real data set about relief times (in minutes) of 20 patients receiving an analgesic is analyzed for illustrative purposes.