Non-recursive digital filtering initialization technique (DFI) is applied to initialize the input data adiabatically as well as diabatically following Lynch and Huang (1992) and Huang and Lynch (1993) for a limited area model to study the impact of DFI on the performance of the model for short-range prediction over Indian region. Both versions of DFI are found to generate fairly well-balanced model initial state and could suppress the high-frequency noise from the forecast at least in the first few hours. The forecast results produced from the two versions of DFI are compared with each other and also with the forecast obtained from the uninitialized input.