Bisection debugging, which is based on binary search over software version history, is widely used in practice to identify the bug introducing commit. However, this process can be expensive because it requires costly compilation and test execution at many commits. We introduce a novel technique--selective bisection (consisting of test selection and commit selection)--to reduce the number of bisection steps, the number of compiler invocations, and the number of executed tests. We evaluated selective bisection on 10 popular open-source projects by performing 25,690 debugging sessions and measuring: (1) savings in number of compiler invocations obtained by commit selection, (2) savings in number of executed tests obtained by test selection, and (3) savings in overall debugging time by selective bisection. Our results show that, in 65% of debugging sessions, commit selection saves between 14% and 71% compiler invocations. Test selection saves 74% of testing effort on average (ranging from 42% to 95%) compared to when developers do not use any test selection. Finally, we demonstrate that one can save substantial time using selective bisection for large projects.