The purpose of the study aims to estimate frequencies of annual maximum 1-day rainfall for ungauged sites in Taiwan using regionalization approach. The index flood method with parameters estimated by L-moments is used to establish the regional frequency model of dimensionless annual maximum 1-day rainfall. Kriging is then employed to estimate the mean annual maximum 1-day rainfall of ungauged sites. Delineation of homogeneous regions is determined by cluster analysis in this study based on the coordinates of the rainfall gauge stations, the means and coefficient of variation of the annual maximum 1-day rainfall. The L-moment based discordancy, heterogeneity, and goodness-of-fit measures are then used to detect unusual sites and select the optimal regional probability models. In this study, a total of 77 rainfall gauge stations are used as the basis to estimate the frequencies of the annual maximum 1-day rainfall for ungauged sites. The number of homogeneous regions derived by the cluster analysis is 3. The best regional probability model for one region is the generalized Pareto distribution, and the Pearson type Ⅲ distribution is the best model for the other two regions. Frequency analysis for ungauged sites needs to establish the variogram models of the mean and coefficient of variation of the annual maximum 1-day rainfall first. The obtained variogram models are then used to estimate the mean annual maximum 1-day rainfall for the ungauged sites. The ungauged sites belong to which homogeneous regions depend on the minimum distance to the centroid of the homogeneous regions. Combined with the derived regional frequency model and estimated mean annual maximum 1-day rainfall, the computing procedures of frequency analysis for ungauged sites are identical with the procedures of gauged sites.