Edge detection is a powerful tool used in geological features identification, such as faults and channels. In this paper, we present a new method for fault detection based on surface fitting algorithm, which is a popular method used in image edge detection. For each point in seismic volume, we will find a small neighborhood in a plane parallel to the local reflector with the help of dip estimation. The data in the neighborhood are then approximated by a bivariate cubic function, called facet model. The local gradient of the function is then being calculated latterly, named facet model attribute. To enhance the robustness of the output attribute and suppress noise, the gradient value are summed over a vertical window and normalized by energy. The dataset used in this paper is part of Netherlands offshore F3 block downloaded on the Opendtect website. To evaluate the performance of our method, we also calculate the dip guided Sobel attribute and variance attribute. By comparing with the two attributes, we found the result of our method is more accurate and shows more details in faults detection.