This study aims to investigate the regional drought frequency for the north, central, south, and east regions in Taiwan. Index flood method associated with L-moment estimation are employed to establish the regional drought frequency distribution for each region and to estimate the return periods for various drought magnitudes and drought duration. In this study, standardized precipitation index (SPI) is used to define droughts. Probability distribution for annual rainfall data is determined first, then followed by transforming the annual rainfall data into SPI. A drought event is defined as the period during which the SPI is below zero. The continuous period of negative SPIs is considered as the drought duration and the cumulative SPIs during the drought duration is considered as the drought magnitude. These two drought characteristics are considered for regional drought frequency analysis in this study. The L-moment based discordancy, heterogeneity, and goodness-of-fit measures are used to detect the unusual sites and select the regional probability distributions of droughts. A total of thirty-five annual rainfall data with record length exceeding 30 years from different regions in Taiwan are selected to illustrate the proposed methodology. The results show that the best data fitted dimensionless regional drought magnitude probability distributions for north, central, south, and east regions are Pearson type III, three-parameter lognormal distribution, generalized Pareto, and Pearson type III distributions, respectively. The best fitted dimensionless regional drought duration probability distributions for north and central regions are Pearson type III distribution, while the generalized Pareto distribution is the best data fitted model for drought duration in south and east regions. The at-site frequencies of drought duration and magnitude for each region can then be determined by the index flood method and used to estimate the return periods of drought duration and magnitude for historical drought events.