DNA-protein recognition plays a central role in gene expression and regulation. Despite increasing structural data on DNA-protein complexes, the molecular mechanism of DNA-protein recognition is not well understood yet, partly because of the considerable extent of redundancy in the base-amino acid interactions as well as of the structural flexibility present within the same interaction pair. To understand the specificity of such interactions, we should examine the interaction energetics by taking account of the structural flexibility. We describe a strategy for elucidating the specificity of DNA-protein interactions by computer simulation, in which free energies of interactions between the amino acid side chains and base pairs are computed by extensive conformational sampling. The simulations enable us to estimate thermodynamic quantities, such as the interaction free energy, enthalpy, and entropy, for each given position of the C-alpha atom of the amino acid side chain by conformational averaging, and to evaluate the free-energy map around base pairs. We report the results for the interactions of Asn with base pairs. In the case of an A-T base pair, we observed a curved valley-shaped region of free-energy minima on the major-groove side of A. Inspection of global-minimum energy configurations of Asn-A interactions in this valley region shows specific double hydrogen bonds, N-H ... N7 and C=O ... HN6, commonly observed in DNA-protein complex structures. On the other hand, we also observed other kinds of double and single hydrogen bonds depending on the C-alpha position in this valley region. Furthermore, the Asn side chain at various positions of C-alpha is flexible enough to readjust its conformation to form similar interactions with the base pair. These results demonstrate the importance of structural flexibility for the specific interactions involved. The free-energy map shows subtle differences from the interaction energy map, revealing the role of entropy in the specificity. The comparison of the free-energy maps for G-C and A-T revealed significant differences in their patterns. These results suggest that our method can be used to dissect the mechanism of base discrimination by amino acids.