Flooding and flash floods, causing loss of life and excessive damage to property, are one of the worst weather-related hazards. Korea is also no exception from the damage. During the last ten years ('93-'02), 129 life losses, 22,320 flood victims, and 1.07 billion dollars property damage on the annual average basis have been occurred. One of the characteristics of such a serious damage is caused by flash flood, which is defined a flood that follows the causative event in a short period of time and often is characterized by a sudden increase in level and velocity of a following water body. Most flash floods occur in streams and small river basins with a drainage area of a few hundred square kilometers or less. Such basins respond rapidly to intense rainfall rates because of steep slopes and impermeable surface, saturated soils, or because of human-induced alterations to the natural drainage. In order to prevent or mitigate such flash flood disasters in a small river basin, it is necessary to develop a flash flood warning system composed of real-time hydrometeorologic data collection system and reliable flood forecasting model. In this sense, the objective of this study is to develope of Korean flash flood forecasting model for Han River basin in Korea. The flash flood guidance (FFG), refers to the volume of rain of a given duration necessary to cause flooding on small streams, can be used as a precaution for potential flash flood. It is generally composed of threshold runoff computation and model-driven soil moisture estimation for hydrologic component and precipitation forecast for meteorological component. In this study the developed results for FFG model on Han River basin are shown and evaluated. For the hydrologic component of FFG model, threshold runoff values are estimated for high-resolution sub-catchments with area is to 49.06km2 on the basis of their delineated geometry, the regional regressions of channel cross-sectional characteristics, and the matching of surface runoff peak to bankfull flow. The results show that the threshold runoff estimates for these small headwater source basins are low (10.09 mm/1-hr~27.58 mm/6-hr), and that if the soil were to be completely saturated in these areas the likelihood of flooding will be high. And, the semi-distributed TOPMODEL is developed for the on-going soil moisture estimates over the Han River basin. Several basic statistics between observed and model computed flows for all selected events are computed for evaluate the model performance. For example, the correlation coefficients between the observed and simulated flows are above 0.90 for the selected events. All these statistics confirm that the model-driven soil moisture can be representative over the study area and can be used as a component of FFG model. The FFG model generates rainfall-runoff curves for each Han-River forecast basin where areal FFG is desired. Using model representations of the current soil moisture, several rainfall values are selected and corresponding runoff values are computed to define rainfall-runoff curves for 1-, 3-, and 6-hour duration. In this process the rainfall that would saturate the soil moisture and overflow the threshold runoff for each duration time is defined as FFG on the Han River sub-catchments. The computed FFG for case study periods during July 2004 and 2006 well rendered sub-catchments very vulnerable to floods. For the one of meteorological component of the FFG model, the stochastic-dynamic bias adjustment procedure for radar rainfall estimation is developed in operational real-time concept and it is used to forecast the flash flood warning. The bias correction scheme is based on the logarithmic gauge-to-radar (G/R) stochastic adjustment technique by using Kalman filtering. The radar data from the composite CAPPI reflectivity and ground rain gauges from KMA is used in this study. The results indicate that radar-derived rainfall estimates from Marshall-Palmar relationship are underestimated than the rainfalls from ground observation for the selected periods. The average G/R ratios are 1.32 and 2.03 for two month events, respectively. Thus the large G/R errors in the radar rainfall estimation are corrected by using the operational adjustment scheme. The average G/R ratios produced by corrected radar rainfall are reduced to 1.09 for July 2004 and to 1.21 for July 2006. And then this well corrected radar rainfall is applied to forecast the flash flood warning in the FFG model. The results show that the corrected radar rainfalls give well forecast flash flood prone area in 1-hour forecast time of the operational use. The forecast rainfall which is the other meteorological component of FFG model for each duration time is provided from RDAPS (5-km) NWP model and VSRF model. It is used to forecast the flash flood watching compared with predicted FFG. For this work FFFT defined to the future flash flood threat are computed in an operational environment. The results show that the FFFT forecasted by RDAPS and VSRF model is not predicted the real flash flood cases for July 2006 in the study area due to qualitative and quantitative errors of forecast rainfall. The other hand the hindcast case study by using the observed rainfall for evaluating an accuracy of the developed Korean FFG model is well forecast the flash flood area in temporal and spatial scale. Therefore, the developed model would be very useful for the real-time forecast of the flash flood warning and watching.