Zonal-scale patterns of precipitation change, as reconstructed for the Mid-Pliocene and two Pleistocene maxima, are compared with those generated in standard 2*CO2-1*CO2 equilibrium experiments by two high-resolution GCMS of equal sensitivities of precipitation and temperature to CO2 doubling. We find that the three warm paleoclimates, despite differences in boundary conditions/forcings, exhibit a similarity in zonal-scale patterns of change for precipitation over land in the Northern Hemisphere (NH); the between-epoch pattern correlation is 0.9 on the average. The two models give marked differences in zonal distribution of precipitation anomalies at mid-latitudes; the between-model pattern correlation for changes of precipitation over NH land is 0.4. The response of precipitation over the NH land area to the NH warming is about 10%/°C in the paleodata compared to 3%/°C in the models. The largest model/paleodata discrepancy refers to the present-day desert belt, where a large precipitation anomaly persists in all epochs. North of 50 N, the absolute values of the zonally-averaged precipitation anomalies simulated by both models fall in the range implied by the three warm paleoclimates, but they are systematically lower than the anomalies of the Mid-Pliocene. If our reconstructions are valid and if climate changes in the Mid-Pliocene were driven solely by CO2 changes, then our results suggest that models are underestimating the magnitude of the precipitation response, especially in the regions of subtropical deserts; the magnitude of the simulated temperature response at high latitudes is also underestimated. At least part of the reported model/paleodata discordance appears to be due to lack of interactive land surface package in the models examined.