The success of any statistical steganalysis algorithm depends on the choice of features extracted and the classifier employed. This paper proposes steganalysis using random forests (SURF) employing HCS (Huffman Code Statistics) features and FR Index (ratio of File size to Resolution). The proposed algorithm is validated over an image database of over 30,000 images spanning various sizes, resolutions, qualities and textures to detect four widely used steganographic schemes namely LSB (Least Significant Bit) encoding, JPHS (JPEG Hide & Seek), MBS (Model Based Steganography) and PQ (Perturbed Quantization). The SURF algorithm proves random forest to be an efficient classifier for steganalysis and its performance is found to be superior compared to current steganalysis methods.