Due to the scale of Taiwan island, the 6-h accumulated rainfall forecast is the most critical information for the water resources management and disaster mitigation during the invasion of typhoon. To improve the forecast, distribution of the 6-h accumulated typhoon rainfall over Taiwan area was investigated through empirical orthogonal function (EOF) analysis. Several forecasting methods were compared. The data set used in the study includes the rainfall of 20 Central Weather Bureau surface stations when typhoons were inside the domain between 18oN and 28oN, 116oE and 126oE from 1961 to 1996. Analysis results show that the first three EOF modes are well separated from the other modes. Those three modes contain about 66% of the total rainfall variances. At some stations, only those three modes can effectively represent the station rainfall. The first EOF mode not only reveals the in-phase increasing or decreasing rainfall on all stations of Taiwan but also shows larger rainfall occurs at stations with higher elevation and larger slope. The enhancement of rainfall on mountainous areas is an indication of topography effect on redistributing rainfall. The second mode and the third mode both show out-of-phase distribution of rainfall over Taiwan when typhoons are nearby. The rainfalls are enhanced on the up-wind sides of the mountain, and the rainfalls are suppressed on the down-wind sides. Those distributions again reveal the effects of redistributing rainfall by Central Mountain Range of Taiwan. To demonstrate how the EOF mode information may be used on the typhoon rainfall forecast, several forecasting methods were evaluated. Those methods include four basic methods and four ensemble forecasts. The basic methods are: Climatology Average Method, Deviation Persistence Method, regression equation forecast based on station rainfall, regression equation forecast based on the amplitude of the first three EOF modes. The ensemble forecasts are the simple averages from the last three basic methods. The results show those basic methods are compatible to each other except Climatology Average Method that underestimates all heavier rains and is outperformed by the others. Since the amplitudes of EOF modes depend on how the rainfall is distributed spatially over all stations of Taiwan, therefore a forecast method based on EOF modes is capable of providing different information from other methods that based on factors only related to a single station. The study also shows that the ensemble forecasts outperform their corresponding member forecasts.