This paper analyzes cross-polarization cancellation in dual-polarization digital radio links transmitting M-ary quadrature amplitude-modulation (M-QAM) signals. We consider the use of three options in the receiver: (1) no cancellation; (2) ideal (i.e., total) cancellation; and (3) optimal nondispersive cancellation. For every option, we assume the canceler to be followed, in each polarization branch, by an ideal minimum mean-square error equalizer. By postulating a statistical model for the co-polarization and cross-polarization responses of a digital radio channel, and then simulating thousands of sets of these responses, we obtain curves that relate outage probability to the number of modulation levels. We show graphically that the no-canceler case is unthinkable; that total cancellation permits results close to those for single-polarization transmission; and that optimal nondispersive cancellation can have a limited range of application. We also examine the effects of key system parameters and the various modeling assumptions.