When using digital camera to get natural images, there exits Color Filter Array (CFA) interpolation effect, which makes great correlation between color components of the images. A new method was proposed to detect splicing operation by using interpolation characteristics of CFA interpolation process. Firstly, the prediction error was obtained by CFA interpolation for image color components. Then, the local weighted variance of image block predicting error was calculated to get the CFA characteristics of the natural image. Finally, to classify and derive the local splicing area, the Gaussian Mixture Model (GMM) was used according to extracted features. Experimental results in the standard splicing tamper the image data set demonstrate that the proposed method can effectively detect the exact location of the tampered area of the image.