Composite elements play an important role in the regulation of transcription. Existing methods for the revealing of potential composite elements are usually based on assessment of the significance of the mutual presence of the predicted transcription factor binding sites using weight matrices or other methods trained on samples of binding sites of known transcription factors. Thus, such methods essentially depend on the completeness of training samples and information on existing TFs. We have proposed a method for de novo discovery of potential composite elements, which does not require preliminary information about the localization of potential TFBS. Using the proposed approach, context signals are identified in the ChIP-Seq dataset, which can correspond to potential composite elements.