We assessed the capability of the regional climate model (RCM) approach in simulating the monthly 50th and 99th percentiles of marine wind speed and significant wave height on the Japanese coast during tropical cyclone season (June-October) of 2002-2004. To simulate the marine winds, we performed dynamical downscaling of the Japanese Reanalysis (JRA-25) using three RCMs. We found that, by applying the Brier skill score, coastal wind speed of the RCMs is closer to the observed, when compared to the JRA-25. The better performance of the RCM approach was found to come from a successful reduction of the bias rather than that of the root mean square error. The RCMs, however, simulate erroneously low (high) marine wind speed northeast of Japan (along the Pacific coast and Kuroshio Extension). The erroneous marine wind speed was found to be related with the sea surface temperature bias associated with the coarse horizontal resolution of the JRA-25. We employed the dynamically-downscaled marine winds to force a wave model (WaveWATCH-III), and found that the marine winds of the RCMs are more effective than the JRA-25 in reproducing significant wave height on the Japan Sea coast.