Soil moisture trends, particularly during the growing season, are an important possible consequence of global warming. Climate model simulations of future soil moisture changes should be made with models that can produce reliable simulations of soil moisture for past climate changes. In this paper, we compare soil moisture simulations from Intergovernmental Panel on Climate Change Fourth Assessment climate models forced with observed climate forcings for the past century, and evaluate them using in situ soil moisture measurements from over 140 stations or districts in midlatitudes of the Northern Hemisphere. To account for the observed spatial scale of soil moisture variations, we used regionally averaged soil moisture for six regions. The models showed realistic seasonal cycles for Ukraine, Russia, and Illinois, but generally poor seasonal cycles for Mongolia and China. To explore the summer drying issue for the second half of the 20th century, we analyzed the linear trend of soil moisture for Ukraine and Russia. Observations from both regions show increases in summer for the period from 1958--1999 that were larger than most trends in the model simulations. Only two out of 25 model realizations show trends comparable to those of observations. These two trends, however, are due to internal model variability rather than a result of external forcing. Changes in precipitation and temperature cannot fully explain soil moisture increases for Ukraine and Russia, which indicates that other factors might have played a dominant role on the observed patterns for soil moisture. We suggest that changes in solar irradiance (the dimming effect) and resultant changes in evaporative demand explain most of the observed soil moisture trends. To understand such sensitivity, we analyzed soil moisture outputs in a special version of the ECHAM5 model that was capable of capturing the observed radiation pattern as a result of incorporating a sophisticated aerosol scheme. Results suggest that both radiation and precipitation patterns are required to be adequately simulated to reproduce the observed soil moisture trends realistically.