Abstract: A heteroscedasticity regression analysis method for censored data is presented,and it extends the traditional regression analysis,which is only suitable to complete and homoscedasticity data,to censored and heteroscedastic data.The heteroscedasticity regression equation is established and the best linear unbiased estimators for the regression parameters,the standard deviation and their covariance are also given.The linear heteroscedasticity regression analysis methods for normal distribution,extreme distribution and Weibull distribution are discussed in detail and the confidence limits of the regression curve and percentiles are given.At present,heteroscedastic data can only be analyzed by group test method that requires large sample.Compared with traditional method,the present method can reach higher precision and a great number of specimens can be saved.