Efficient filtering technique on removal of Gaussian noise is not an easy task. In this paper two same centered different size windows are taken for every pixel of the noisy image. A variable threshold value (Γ) has been calculated with the help of standard deviation (σ) of the pixels in the big window. The pixels of the big window are distributed in four sets to determine the conditional mean. Finally an eight bit binary table is created for decision and restoration. Results of the computer simulations demonstrate the effectiveness of the proposed method taking different test images at variable noise densities.