This study evaluated the downscaled downward shortwave radiation (S_d) and daily mean, maximum, and minimum surface air temperatures (T_m, T_x, and T_n) over Japan derived from four dynamical models and one statistical model for the period 1985-2004. These variables, S_d, T_m, T_x, and T_n, are often used as input data for impact models, such as crop growth models. Therefore, the evaluation of these variables is essential prior to the application of the downscaled climate data to impact model simulation. All models except for the statistical model remarkably overestimated the S_d throughout the year, whereas the area-averaged seasonal change of the temperature variables was accurately simulated. On the other hand, the statistical model accurately simulated the area-averaged seasonal change of S_d and temperatures. These tendencies were also observed for the area- and time-averaged values. The inconsistency found for dynamical models in the relationship between the S_d bias and each of the temperature variable biases is attributed to the large inter-model differences, such as the difference of the radiation scheme, the sub-grid scale cloud parameterization, and the land surface scheme.