In this paper, we present a steganalytic method to attack JPEG2000 lazy-mode steganography proposed by Su et al. The key element of the method is the Hilbert-Huang transform based analysis of the code-block noise variance sequences of stego images and non-stego noisy images. The Hilbert transform based characteristic vectors are constructed via empirical mode decomposition of the sequences and the support vector machine classifier is used in classification. Experimental results have demonstrated effectiveness of the proposed steganalytic method. According to our best knowledge, this method is the first successful attack of JPEG2000 lazy-mode steganography. And furthermore, the proposed method takes first step towards the application of Hilbert-Huang transform in steganalysis and proves its great advantage.