This paper builds up a pattern recognition system to detect anomalies in JPEG images, especially steganographic content. The system consists of feature generation, feature ranking and selection, feature extraction, and pattern classification. These processes tend to capture image characteristics, reduce the problem dimensionality, eliminate the noise inferences between features, and further improve classification accuracies on clean and steganography JPEG images. Based on the discussion and analysis of six popular JPEG steganography methods, the entire recognition system results in higher classification accuracies between clean and steganography classes compared to merely using individual feature subset for JPEG steganography detection. The strength of feature combination and preprocessing has been integrated even when a small amount of information is embedded. The work demonstrated in this paper is extensible and can be improved by integrating various new and current techniques. © 2012 Elsevier B.V. All rights reserved.