Abstract: Traditional fault detection methods are based on the singularity among traces; the fault and noise are being detected at the same time. If large-scale fault detecting methods are adopted or noise suppression methods used together with fault detection methods, the signal-to-noise ratio will increase, resulting in fuzzy fault boundaries. Therefore, signal-to-noise ratio and detecting precision are incompatible during fault detection. In order to solve this problem, an improved method based on direction information measurement is proposed to suit the situation of non-catastrophe for signals outside the boundary, considering that directivity is one of the essential differences between fault and noise. This method in combination with the wavelet small-scale edge detecting method can effectively detect fault and suppress noise. The practical application shows that the imaging of fault is clear and the correlation of fault direction and structure stress is strong, which approved the validity of the method.