Observation data and output of a regional climate model, RegCM2.5, nested in a global climate model, NCAR-CSM, were analyzed to determine the standard deviation and frequency of extreme values, which were used as indices to characterize the variability of the daily minimum ðT min Þ and maximum ðT max Þ temperatures in Japan. The frequency of extreme values, which are defined in this paper as deviations greater than twofold the standard deviation, was estimated from data using a new method. The significant results of this analysis of observation data are as follows. With regards to both T min and T max , the significant seasonal variations of standard deviation are the annual and semiannual ones, with the minimum value in August and a weak minimum in January. The region where the standard deviation is large drifts northward from February to May, with a maximum standard deviation in April. The central latitude almost agrees with that where the zonal wind is maximum at 500 hPa. As compared with the normal distribution, the extremely high temperature in summer, and the extremely low temperature in winter, are observed less frequently. The indices calculated from the model's output after simulating the current climatic condition were compared with the observation data for evaluation. From this comparison, it is shown that the variability characteristic of T max was not reproduced well; however, that of T min was reproduced well by the model. Therefore, it is concluded that T min reproduced by RegCM/CSM can be used to predict the change in the frequency of extreme values by correcting the systematic temperature bias, which is within G1 K.