BACKGROUND: Offering patients in oncology trials the opportunity to cross over to active treatment at disease progression is a commonly used strategy to address ethical issues associated with the use of placebo controls, but could lead to statistical challenges for the analysis of key endpoints such as overall survival. While an advantage from the perspective of the treated patient enrolled in the trial, cross-over leads to loss of information and dilution of the comparative clinical efficacy and cost effectiveness results. OBJECTIVES:  The purpose of the study is to compare alternative methods for analyzing overall survival data in the presence of cross-over, thus illustrating differences between methods, and providing guidance on choice of methodology. METHODS:  Two promising methods for dealing with cross-over are inverse probability of censoring weighting, and the rank-preserving structural failure time model. The methods are compared with naïve censoring of data at cross-over and intention-to-treat analysis ignoring cross-over using two recent examples of trials in oncology: the receptor tyrosine kinase inhibitor sunitinib in renal cell carcinoma (RCC), and in gastrointestinal stromal tumor (GIST). RESULTS:  The analyses showed that for a trial with a low proportion of cross-over from placebo to active treatment (RCC), the choice of statistical method did not affect the results to a great extent; the range of relative mortality risk for active treatment vs. control was narrow. With a high proportion of cross-over (GIST), the range of relative mortality risks was broader. CONCLUSIONS:  Naïve censoring at cross-over can lead to bias and should be avoided. If cross-over occurs frequently, the inverse probability of censoring weighting method or the rank-preserving structural failure time model are recommended depending on the characteristics of cross over in the trial, trial size and available data.