Some neurological patients make more errors naming living than nonliving things. This is true even when the difficulty of naming living and nonliving things, as measured by the performance of normal subjects, is taken into account. Gaffan and Heywood (1993) argued that this apparently selective deficit for living things is spurious and offered an alternative explanation of it in terms of the effect of noise on regression analyses. We report Monte Carlo simulations showing that, in two patients, the living things deficit cannot be accounted for this in this way. These simulations are more consistent with the claim that the recognition of living things depends on specialized mechanisms.