The classical differential subordination of martingales, introduced by Burkholder in the eighties, is generalized to the noncommutative setting. Working under this domination, we establish the strong-type inequalities with the constants of optimal order as p -> 1 and p -> infinity, and the corresponding endpoint weak-type (1, 1) estimate. In contrast to the classical case, we need to introduce two different versions of noncommutative differential subordination, depending on the range of the exponents. For the L-P-estimate, 2 <= p < infinity, a certain weaker version is sufficient; on the other hand, the strong-type (p,p) inequality, 1 < p < 2, and the weak-type (1, 1) estimate require a stronger version. As an application, we present a new proof of noncommutative Burkholder-Gundy inequalities. The main technical advance is a noncommutative version of the good lambda-inequality and a certain summation argument. We expect that these techniques will be useful in other situation as well. (C) 2018 Elsevier Inc. All rights reserved.