DNA bending is believed to facilitate the initial recognition of the mismatched base for repair. The repair efficiencies are dependent on both the mismatch type and neighboring nucleotide sequence. We have studied bending of several DNA duplexes containing canonical matches: A:T and G:C; various mismatches: A:A, A:C, G:A, G:G, G:T, C:C, C:T, and T:T; and a bis-abasic site: X:X. Free-energy profiles were generated for DNA bending using umbrella sampling. The highest energetic cost associated with DNA bending is observed for canonical matches while bending free energies are lower in the presence of mismatches, with the lowest value for the abasic site. In all of the sequences, DNA duplexes bend toward the major groove with widening of the minor groove. For homoduplexes, DNA bending is observed to occur via smooth deformations, whereas for heteroduplexes, kinks are observed at the mismatch site during strong bending. In general, pyrimidine:pyrimidine mismatches are the most destabilizing, while purine:purine mismatches lead to intermediate destabilization, and purine:pyrimidine mismatches are the least destabilizing. The ease of bending is partially correlated with the binding affinity of MutS to the mismatch pairs and subsequent repair efficiencies, indicating that intrinsic DNA bending propensities are a key factor of mismatch recognition.