A flexible statistical modeling framework for the analysis of creep rupture data is proposed, which offers an improvement on traditional methods of deriving creep rupture strength values and confidence limits. The paper reviews a family of models that can be used to represent the trend relationship between failure times about the trend line, and examines the reliability of extrapolations. Areas of statistical research which would lead to model improvement are discussed, such as variance heterogeneity, left censoring and allowance for the cluster (cast) structure of the data.