The existing reflection components separation methods can not remove the specularity of natural images without estimating illumination chromaticity. To solve this problem, a cross channel gradient coherence based reflection components separation method was proposed. Firstly, homogeneous specular pixels, which consist of major portion of specular regions, were detected by using cross channel gradient coherence, and the reflection components separation tensor was constructed. Then, the reflection components separation tensor was applied to transform the image gradient, which led to the separation of the specular reflection gradient and the diffuse reflection gradient. Finally, the estimated diffuse reflection component could be reconstructed from the diffuse reflection gradient. Experiment results show that the proposed method separates reflection components for natural images without color constancy preprocessing, simplifies the specularity removal process for natural images, and removes specularity for normalized images more accurately.