We constructed a non-redundant non-ribosomal protein-RNA interface dataset (including 694 structures) from the Protein Data Bank (PDB). The interface preferences of amino acids, nucleotides and the secondary structure elements of protein and RNA were computed based on the dataset. The results show that β-ladder, β-bridge and 310-helix of proteins and the unpaired nucleotides of RNA, especially those irregularly arranged nucleotides have remarkably high interface propensities. Based on these, we classified the secondary structure elements, constructed the 60×12 amino acid-nucleotide pairwise potential, and used it as a scoring function in protein-RNA docking to select the near native structures. The results show the 60×12 pairwise potential has a scoring success rate of 65.77%, better than those of the pairwise potentials with secondary structure information of protein or RNA considered, as well as better than that of our previously constructed 60×8* potential based on the 251 protein-RNA complex structures. This work is helpful for strengthening the understanding of protein-RNA specific interactions and can advance the progress of protein-RNA complex structure prediction.