Flash flood is one of the devastating natural disaster in the mountainous region of India. In Sikkim-Darjeeling Himalaya Teesta Watershed is a probable flash flood occurrence zone. The present study is an attempt to develop a simulation model of surface runoff in upper Teesta basin, which directly related to catastrophic flood happenings. At the first stage, a primary unit hydrograph is developed with time for an excess rainfall event by estimating the stream flow response at the outlet of the watershed. Specifically, the methodology was based on meteorological and morphological data processing in the geospatial environment and on data editing. Rainfall time-series data collected from India Meteorological Department and processed to calculate water flow to evaluate peak Discharge over time and estimate the runoff volume. Apart from the meteorological data, background data such as topography, drainage network, land cover and geological data were also collected. Clipping off the watershed from the entire area and the streamline generation for Teesta watershed was done from Aster DEM data using the LPS Tool and Arc GIS environment. A part of this model is to build geospatially input CN grid data based on land use, hydrological soil group and eluviation of the area. The analysis of the different hydraulic model to detect flash flood probability was done using HEC-RAS, and HEC-HMS Software, which was of great importance to achieve the final result. 313 km2 area was found to be most vulnerable to flash flood includes Singtam, Melli, Jourthang, Chungthang and Lachung and 655 km2 as moderately susceptible includes Teesta Bazar, Rangpo, Yumthang, Dambung and Thangu Valley. The simulation model based on the watershed and its surface characteristics of the soil, land use land cover and altitude of the area and the discharge volume of water with time. This model validated by inserting the rainfall data of a flood event, which took place on 2nd to 5th October in 1968, and 78 % of the actual area flooded reflected in the output of the model.