The paper presents a comprehensive refitting of the Amber ff99SB force field to improve the accuracy of protein side chain and backbone parameters. The authors performed a complete re-fitting of all amino acid side chain dihedral parameters, which were previously derived from ff94. The training set included multidimensional dihedral scans designed to enhance the transferability of the parameters. The results showed significant improvements in all amino acids compared to ff99SB, with average errors in relative energies of pairs of conformations under 1.0 kcal/mol as compared to quantum mechanics (QM) calculations, a 35% reduction from ff99SB. Additionally, empirical adjustments were made to the protein backbone dihedral parameters, specifically to the φ rotational profile, to compensate for the lack of QM training data in the β-ppII transition region. These modifications, collectively called ff14SB, improved the reproduction of benchmarks, secondary structure content in small peptides, and NMR γ1 scalar coupling measurements for proteins in solution. The paper also discusses the Amber ff12SB parameter set, a preliminary version of ff14SB that includes most of its improvements. The authors recommend the use of ff14SB for protein simulations in Amber and other biomolecular simulation software.The paper presents a comprehensive refitting of the Amber ff99SB force field to improve the accuracy of protein side chain and backbone parameters. The authors performed a complete re-fitting of all amino acid side chain dihedral parameters, which were previously derived from ff94. The training set included multidimensional dihedral scans designed to enhance the transferability of the parameters. The results showed significant improvements in all amino acids compared to ff99SB, with average errors in relative energies of pairs of conformations under 1.0 kcal/mol as compared to quantum mechanics (QM) calculations, a 35% reduction from ff99SB. Additionally, empirical adjustments were made to the protein backbone dihedral parameters, specifically to the φ rotational profile, to compensate for the lack of QM training data in the β-ppII transition region. These modifications, collectively called ff14SB, improved the reproduction of benchmarks, secondary structure content in small peptides, and NMR γ1 scalar coupling measurements for proteins in solution. The paper also discusses the Amber ff12SB parameter set, a preliminary version of ff14SB that includes most of its improvements. The authors recommend the use of ff14SB for protein simulations in Amber and other biomolecular simulation software.