We propose a model to describe the frequencies of site-specific base substitution errors by DNA polymerase. In the model, nucleotide misinsertion frequencies are determined by 5'-nearest-neighbor base stacking and 3'-exonucleolytic proofreading efficiencies are governed by the relative proportions of G . C base pairs in the region surrounding the misinserted nucleotide. The model is used to analyze the frequency of replacing dAMP by 2-aminopurine deoxyribonucleotide with purified bacteriophage T4 L141 antimutator DNA polymerase at 57 sites on phi X174 DNA (Pless, R. C., and Bessman, M.J. (1983) Biochemistry 22, 4905-4915). A linear least-squares fit yields a correlation coefficient of 0.83 and a standard deviation of 2.8% between predicted and observed results. Four to five base pairs on each side of the 2-aminopurine incorporation site, approximately one double-helical turn, are found to exert a maximal influence on proofreading. Thermal melting data on native and synthetic DNA are used to deduce base-stacking energies for nearest-neighbor doublets including those involving 2-aminopurine. The inclusion of base-stacking energies in the model calculations leads to predictions similar to those based on the use of empirical parameters for individual base pairs.