In this paper, a new method for defective region detection of wood surface knot using color and texture information was proposed. At first, initial pixels were selected automatically from defective regions based on color histogram through the differences between defective regions and normal wood regions. Then a new algorithm called texture diffusion was proposed to search the contours of defective regions from the initial pixels based on local texture information. Besides, in order to suit the process of texture diffusion, a modified LBP operator named LBP-TD operator was also proposed in this new algorithm. For several kinds of wood surface knot defects, the experiments showed that when obvious differences of color and texture information between defective regions and normal wood regions existed, the accurate contours of knot defective regions could be obtained by the proposed method whether the normal wood surface texture was regular or not. When the differences of color and texture information between defective regions and normal wood regions were not obvious, approximate contours of knot defective regions can also be obtained. Besides, the experimental data showed that the error detection rate of the proposed method was lower than that of OTSU.