In this study, the precipitable water (PW) and ice water path (IWP) simulated by the Global Data Assimilation System (GDAS) are compared to those observed by NOAA's Microwave Surface and Precipitation Products System. Results show small root-mean-square (RMS) differences in PW but large RMS differences in IWP between the two data sets, indicating the existence of model errors in reproducing clouds. To examine the possible linkage between the small PW and large IWP differences, three experiments are conducted with a two-dimensional cloud-resolving model in which the observed zonal wind and the GDAS-derived large-scale vertical velocity are imposed. The model initial conditions of PW are perturbed by ¿10% in the first two experiments, respectively, while treating the third one without any perturbation as a control simulation. Thermodynamic, cloud microphysics, and precipitation budgets are then calculated from the zonally averaged and vertically integrated data at hourly intervals from these experiments. Results show the generation of larger differences in the cloud hydrometeors and surface rain rates, with the given PW perturbations. This indicates that the model-simulated clouds and precipitation are extremely sensitive to the initial errors in PW, primarily through the biased condensation process.