Abstract: In this study, texture edge detection using a combined statistical and structural approach based texture representation is presented. Observable textures can be characterized by a set of primitives and their placement rules. Accordingly, a set of texture primitives are proposed for the representation of small region and the entire texture image is considered to be consisting of these proposed set of primitives. The distribution of these primitives is obtained as the global descriptor for the texture image, namely, the texture primitive spectrum. In this study, the usage of the texture primitive spectrum has been shown effective for the texture edge detection. Texture edge detection is defined as segmenting the texture image into mutually exclusive regions. The edge detection has been attempted for 2 different class of texture images, namely, deterministic (texture region is consisting of a single primitives distributed over the entire region) type of textures and non deterministic type of textures. In the first category, the texture primitives are replaced with the corresponding primitive numbers and conventional edge detection is performed on these array of numbers to detect the edges. For the second category, texture descriptors are obtained for a larger window size and conventional edge detection with these descriptor is performed and the edges are detected.With all these experiments and results, the usage of the proposed descriptors are effective and are promising.