At site analysis is not appropriate if the record length is shorter than the target return period T. If the record length is longer than 2T years, then at-site analysis may be sufficient. However, in such a case, regional analysis is recommended for purpose of comparison. Sample sizes of annual maximum rainfall data in Korea are usually shorter than 50 years. Therefore, it is essential to use regional frequency analysis for estimating rainfall quantiles of more than 100 years return period. In this research, regional rainfall frequency analysis is performed for hourly rainfall data of South Korea. The identification of homogeneous regions is very important and requires subjective judgments in regional frequency analysis. Cluster analysis is used to form homogeneous regions for regional frequency analysis. An appropriate distribution is chosen by goodness-of-fit test. GLO is found to be an appropriate distribution as a result of goodness-of-fit measure. Simulation experiments are performed to check the performance of frequency analysis techniques. The effects of discordant sites on quantiles are considered. Copyright ASCE 2005.