We propose a model for the kinetics of reversible addition-fragmentation chain transfer (RAFT) polymerization. The essence of this model is that the termination of the radical intermediate formed by the RAFT process occurs only with the shortest active radicals. This model accounts for the absence of 3-armed stars predicted by other cross-termination models since the short radical makes a negligible difference to the overall molecular weight. The model is tested against experiments on styrene at 60 degrees C with cyanoisopropyl dithiobezoate (CPDB) as the RAFT agent. The predicted rate coefficients are consistent with slow fragmentation of the RAFT intermediate, and the overall concentration of radicals is consistent with ESR experiments. Overall, it demonstrates that the two conflicting models that have been proposed so far can actually coexist.