Subhalo abundance matching (SHAM) is a technique for populating simulated dark matter distributions with galaxies, assuming a monotonic relation between a galaxy's stellar mass or luminosity and the mass of its parent dark matter halo or subhalo. We examine the accuracy of SHAM in two cosmological SPH simulations, one of which includes momentum-driven winds. The SPH simulations indeed show a nearly monotonic relation between stellar mass and halo mass provided that, for satellite galaxies, we use the mass of the subhalo at the epoch when it became a satellite. In each simulation, the median relation for central and satellite galaxies is nearly identical, though a somewhat larger fraction of satellites are outliers. SHAM-assigned masses (at z=0-2), luminosities (R-band at z=0), or star formation rates (at z=2) have a 68% scatter of 0.09-0.15 dex relative to the true simulation values. When we apply SHAM to the subhalo population of collisionless N-body simulation with the same initial conditions as the SPH runs, we find generally good agreement for the halo occupation distributions and halo radial profiles of galaxy samples defined by thresholds in stellar mass. However, because a small fraction of SPH galaxies suffer severe stellar mass loss after becoming satellites, SHAM slightly overpopulates high mass halos; this effect is more significant for the wind simulation, which produces galaxies that are less massive and more fragile. SHAM recovers the two-point correlation function of the SPH galaxies in the no-wind simulation to better than 10% at scales 0.1 < r < 10 Mpc/h. For the wind simulation, agreement is better than 15% at r > 2 Mpc/h, but overpopulation of massive halos increases the correlation function by a factor of ~2.5 on small scales.