Gaugecorrected terrestrial precipitation records and central Arctic Ocean data from the Russian "North Pole" series of drifting stations are used to examine the ability of the NCEP/NCAR and ERA reanalysis models to simulate interannual variability in Arctic precipitation. Results are based primarily on the period 19791988. Both models capture the mean spatial distribution of annual and monthly Arctic precipitation but with considerable error in magnitude. Model performance is best during winter and worst during summer. A significant problem seen only in the NCEP/NCAR model is severe oversimulation of summer precipitation over land areas, due to excessive convective precipitation. This is most apparent over Eurasia, where the model consistently depicts the precipitation maximum as occurring during July, one month too early. In turn, unrealistically high evaporation rates show that the NCEP boundary layer is too wet. The ERA model correctly captures variability in the timing of the seasonal precipitation peak, and better simulates variability in seasonal and annual totals. The problem with the NCEP model could arise from problems in the analyses, model physics or a combination of both and we are unable to identify a specific cause. Existing studies suggest that part of the problem lies in the convective parameterization itself. Alternatively, it is reasonable to suspect that the convective precipitation problem with the NCEP model relates at least in part to excessive boundary layer heating due to the high radiation fluxes. Note that this problem would not give rise to the convective problem over sea ice cover as the surface temperature is not free to rise and sea ice extent is initialized to observed distributions. It is not immediately clear, however, why excessive convective precipitation over land occurs in association with limited cloud cover. Another possible culprit is soil moisture, which NCEP updates by the modeled precipitation. If soil moisture is too high, this would favor excessive evaporation and precipitation, which in turn feeds back to keep soil moisture high. By comparison, the ERA soil moisture is initialized from analysis increments in the lowlevel humidity, which constrains the values. Although the NCEP downwelling shortwave fluxes are much too high, which could also contribute to excessive evaporation, this may be compensated for by positive biases in albedo. The performance of the NCEP model improves over the Canadian Arctic Archipelago, where summer convective precipitation is suppressed due to the large areas of cold open water and sea ice. However, both models have problems in simulating precipitation over the icecovered central Arctic Ocean, indicative of model deficiencies other than those associated with convective precipitation. Although notable differences are clearly evident, the ERA modeled data in particular appears sufficiently realistic to represent a base for blending with other data to provide gridded fields for climate research and evaluation of climate and sea ice models