In many areas of clinical settings, risk models are often developed to predict a patients’ future health outcomes and thereafter to guide clinical management of patients. To appreciate the quality of these models in clinical predictions, it is important to evaluate their predictive ability. Measures to assess the predictive performance of models for binary data are reasonably well developed. Despite the proposal of several measures for survival risk models, it is still unclear which measures should be generally used in practice. This talk focuses on an investigation of a wide range of measures using a simulation study in order to make some practical recommendations. Measures were evaluated with respect to their robustness to censoring and their sensitivity to the omission of important predictors. The simulation results show that a few measures under investigation performed well in a range of scenarios while the other measures had poor performance particularly in the presence of censored data. Based on the simulation study, some recommendations for using these measures in practice are discussed.