SWAT is a physically explicit distributed hydrologic model, which could simulate hydrological processes such as surface runoff, snowmelt runoff and infiltration with geographic information system(GIS).The surface runoff component in SWAT is implemented by using the SCS curve number and snowmelt is calculated by a relatively simple, empirical degree-day method. Those approaches work well in humid and semi-humid areas where precipitation dominantly controls runoff. In cold and arid regions, for example, in a case study in the Heihe river basin (HRB) of northwest China, however, it revealed those implementations cannot represent the effects of snowmelt in springs and thus impact surface runoff simulations. At large, snowmelt in such areas is underestimated and more discrepancies are consequently introduced to the overall simulation accuracy. The FASST model is a surface process model with explicit physical base, including a snowmelt runoff component that makes use of mass and energy balance equations. There is a snowmelt implementation in FASST, which takes topography, vegetation, soil type, and snow physical properties into account. Existing applications of FASST show good simulations of snowmelt in terrain-complex mountainous watershed and its applicability is also confirmed by an application to the Tangula site of northwest China located in a similar cold and alpine area. This paper proposes a coupling approach to improve the simulation of snowmelt by integrating FASST snowmelt to SWAT. In this approach, when the snowmelt begins to be calculated in SWAT, it will call FASST snowmelt to calculate, return its value to the SWAT corresponding variable and continue remainder SWAT logics. The technical implementation is presented in detail. An application to the upper mountainous HRB is set up to test its performance. There are abundant snow falls in Upper HRB in winters and snowmelt is the primary water source to river in springs that cannot be simulated well by original SWAT. The improvements in comparison to the original were examined from three aspects, namely, streamflow, snowmelt runoff and surface runoff contribution to streamflow. Both snowmelt and surface runoff estimates with the coupled model were increased so that streamflow estimate was more close to that observed. An improvement of monthly streamflow estimation by 0.11 in Nash-Sutcliffe coefficient (NSE) can be achieved. Examination of surface runoff contribution to streamflow also supports its feasibility in estimating snowmelt by the coupled approach. Moreover, by using optimized parameter sets, the monthly streamflow simulation accuracy in the validation period of 2000-2009 can be up to 0.83 in NSE. The results confirm the applicability of the SWAT-FASST coupled approach in cold and alpine watersheds where snowmelt should be taken into account and suggest its significance in improving the simulations in such areas.