The accuracy with which Escherichia coli DNA polymerase I (Pol I) copies natural DNA in vitro has been determined. When phi X174 viral DNA containing an amber mutation (am3) is primed with a single restriction endonuclease fragment, copied in vitro with Pol I and then expressed in E. coli spheroplasts (Weymout, L. A., and Loeb, L. A. (1978) Proc. Natl. Acad. Sci. U. S. A. 75, 1924), the reversion frequency of this DNA is greater than that of uncopied DNA. This change in reversion frequency can be increased by selectively increasing the concentration of either dATP or dCTP relative to the other deoxyribonucleotide substrates. DNA sequence analyses of revertants obtained from substrate pool bias experiments demonstrates that the revertants contain the selectively biased nucleotide as an incorrect substitution at position 587 of the am3 codon. We have analyzed the product of the in vitro Pol I reaction using neutral and alkaline sucrose gradients. Fifty per cent of the input phi X174 DNA template molecules are copied past the am3 site. The phenotypic expression of the product (revertant) strand in the spheroplast assay was estimated using a model heteroduplex molecule similar in structure to the product of the reaction and containing a single base mismatch (A:A or A:C) at position 587. Using these data, and by extrapolation from pool bias experiments, we estimate the error rate of Pol I in Mg2+-activated reactions using equimolar concentrations of the four deoxynucleotide substrates is 1/680,000 for an A:C mispair and < 1/6,300,000 for an A:A mispair at position 587 of the am3 codon in phi X174 DNA.