The fifth-generation PSU-NCAR Mesoscale Model (MM5)-based regional climate model (CMM5) capability in simulating the U.S. soil temperature and soil moisture annual cycle and interannual variability is evaluated by comparing the 1982--2002 continuous integration driven by the NCEP-DOE AMIP II reanalysis (R-2) with observations, the R-2 derivatives and North American Land Data Assimilation System (NLDAS) products. For the annual cycle, the CMM5 produces more realistic regional details and overall smaller biases than the driving R-2 and NLDAS outputs. The CMM5 also faithfully simulates interannual variations of soil temperature over the central United States and soil moisture in Illinois and Iowa, where observational data are available. The existing CMM5 differences from observations in soil temperature (moisture) cannot be fully explained by model biases in surface air temperature (precipitation). Inconsistencies between measurements taken under short grass versus model representations beneath other land cover types may play an important role. In particular, such measurements overestimate soil temperature in summer and fall while generating a 1-month phase lead in the soil moisture annual cycle with respect to croplands in the model. The result emphasizes the need for more comprehensive study on model evaluation and bias understanding of soil temperature and soil moisture.