Four regional climate models (RegCM2, MM5v2/BATS, MM5v2/SHEELS, MM5v3/OSU) were run for two years (1987 and 1988) on a domain centered on the FIFE site in Kansas. All model runs had a horizontal resolution of 20 km and were forced at the lateral boundaries with either ECMWF or NCEP reanalayses. The analysis we perform expands upon previous studies such as PILPS by allowing the land surface schemes to dynamically interact with their native atmospheric components without having various aspects of these interactions arbitrarily prescribed (as was done in PILPS). Comparison of results with observations collected during FIFE indicates all the models have difficulty reproducing various surface variables, though it should be noted that the observations themselves are occasionally suspect. All the models display greater interannual variation than is present in the observations. The study indicates that the more recently developed models are not demonstrably better than the older models.