Rank Preserving Structural Failure Time (RPSFT) and Inverse Probability of Censoring Weighted (IPCW) models are increasingly applied to adjust for bias in treatment effect estimates in trials of cancer drugs where crossover to the new treatment is allowed.  We review the assumptions and processes underlying these methods and their suitability in different situations using a case study.  A first consideration is the meaning of the cross-over adjusted result, which is an estimate of the treatment effect had crossover not occurred – i.e., if patients had continued to be treated with standard therapy.  This may be plausible when no other treatment options are available; otherwise, the crossover-adjusted estimate becomes a theoretical upper-bound, and likely not representative of real-world effectiveness.  The assumptions inherent to each approach must also be considered carefully.  For instance, RPSFT models shrink the survival time of patients who cross-over by a magnitude proportional to the time spent on the experimental drug and its associated benefit.  Implied here is that patients derive the same benefit for every unit of time on the experimental treatment, and that this benefit is the same in both the original randomized population and those who cross over.   The latter are selected, however, based on their expected responsiveness to and tolerance of the new treatment.  In the IPCW approach adjustment is made by analytically matching patients who cross over to patients who had a similar prognosis but did not crossover.  The prognosis of these groups may not overlap sufficiently, however; in fact, full characterization of the prognosis of patients may be impossible as it requires time-dependent information on factors considered by physicians deciding which patients should cross over.   This talk aims to promote better understanding of these methods, their proper application and interpretation and use of results.