Development of new derivatives and analogs of nucleic acids (NA) with precalculated physico-chemical properties is important, both in practice and basic research. Unfortunately, these studies remain laborious, costly, and time-consuming. I silico research probably could resolve this problem due to the significant progress in development of computer software and hardware. The aim of this work is to study a molecular dynamics approach for nucleic acid hybridization enthalpy calculation. The enthalpies of DNA duplex formation were calculated as a difference of the total internal energy of double- and single-stranded states which were averaged from a 10 ns MD trajectory computed with Amber 11 software (UCSF, USA). Computations were performed on NVIDIA GTX580/Intel i7-2600 hardware and resources of Siberian the Supercomputer Center (ICMMG SB RAS). The use of a GPU has speeded up the modeling in implicit solvent up to 60 times and up to 30 times in explicit solvent in comparison with the one node of a CPU. To determine the optimal parameter set of modeling, we have used Dickerson–Drew dodecamer (DDD) 5'-CGCGAATTCGCG-3' with well characterized secondary structure and thermal stability. We have varied force field, temperature, heating protocol, and ion concentration in implicit and explicit solvent, solvent shell radius and compared averaged structures with those experimentally obtained. Using optimal parameters of modeling, we have shown that hybridization enthalpy of DDD correlates well with experimental and calculated ones of the nearest neighbor models’ enthalpies (Lomzov et.al., 2006). The differences were <15% whereas the experimental accuracy is about 10%. To verify the MD predictive ability, we have collected a database of experimentally determined thermodynamic parameters (enthalpy and entropy) of hybridization of 272 oligodeoxyribonucleotides. The length of oligonucleotides varies from 4 up to 16 base pairs (aver. 9 bp), GC-content 0-100% (aver. 57%). The total energy of oligonucleotide or duplex was averaged over 10 000 snapshots of 10 ns trajectories simulated with the optimal parameter set. The correlation between the values of hybridization enthalpies obtained experimentally and calculated using MD are shown in the figure. The RMSD and average error values of calculated and experimental enthalpies were less than 12 and 15%, respectively. The results obtained show that MD modeling allows one to calculate enthalpy of matched DNA duplexes with surprisingly good accuracy.