Despite considerable efforts, structural prediction of protein-peptide complexes is still a very challenging task, mainly due to two reasons: high flexibility of the peptides and transient character of their interactions with proteins. Recently we have developed an automated web server CABS-dock (http://biocomp.chem.uw.edu.pl/CABSdock), which conducts flexible protein-peptide docking without any knowledge about the binding site. Our method allows for full flexibility of the peptide, whereas the flexibility of the receptor is restricted to near native conformations considering the main chain, and full flexibility of the side chains. Performance of the CABS-dock server was thoroughly tested on a benchmark of 171 test cases, both bound and unbound. Evaluation of the obtained results showed overall good performance of the method, especially that no information of the binding site was used. From unsuccessful experiments we learned that the accuracy of docking might be significantly improved, if only little information of the binding site was considered. In fact, in real-life applications user typically has access to some data indicating the location and/or structure of the binding site. In the current work, we test and demonstrate the performance of the CABS-dock server with two new features. The first one allows to utilize the knowledge about receptor residue(s) constituting the binding site, and the second one allows to enforce the desired secondary structure on the peptide structure. Based on the given example, we observe significant improvement of the docking accuracy in comparison to the default CABS-dock mode.