Two-dimensional (2-D) adaptive digital filters (ADF's) for 2-D signal processing have become a fascinating area of the adaptive signal processing. However, conventional 2-D FIR ADF's require a lot of computations. For example, the TDLMS requires N-2 multiplications per pixel. We propose a new 2-D adaptive filter using the FFT's. The proposed adaptive filter has parallel structure, thus we can reduce the computational complexity to O(log(2)N) per pixel.