Retrospective sampling designs, including case-cohort and case-control designs, are commonly used for failure time data in the presence of censoring. In this talk, we propose a new retrospective sampling design, called end-point sampling, which improves the efficiency of the case-cohort and case-control designs. Under different assumptions, the maximum likelihood approach as well as the inverse probability weighting approach is developed respectively to estimate the regression parameters. The resulting estimators are proved to be consistent and asymptotically normal. Simulation and real data studies show favorable evidence for the proposed design in comparison with the existing ones.