Current models suggest that two or more DNA polymerases may be required for high-fidelity semiconservative DNA replication in eukaryotic cells. In the present study, we directly compare the fidelity of SV40 origin-dependent DNA replication in human cell extracts to the fidelity of mammalian DNA polymerases alpha, delta, and epsilon using lacZ alpha of M13mp2 as a reporter gene. Their fidelity, in decreasing order, is replication greater than or equal to pol epsilon greater than pol delta greater than pol alpha. DNA sequence analysis of mutants derived from extract reactions suggests that replication is accurate when considering single-base substitutions, single-base frameshifts, and larger deletions. The exonuclease-containing calf thymus DNA polymerase epsilon is also highly accurate. When high concentrations of deoxynucleoside triphosphates and deoxyguanosine monophosphate are included in the pol epsilon reaction, both base substitution and frameshift error rates increase. This response suggests that exonucleolytic proofreading contributes to the high base substitution and frameshift fidelity. Exonuclease-containing calf thymus DNA polymerase delta, which requires proliferating cell nuclear antigen for efficient synthesis, is significantly less accurate than pol epsilon. In contrast to pol epsilon, pol delta generates errors during synthesis at a relatively modest concentration of deoxynucleoside triphosphates (100 microM), and the error rate did not increase upon addition of adenosine monophosphate. Thus, we are as yet unable to demonstrate that exonucleolytic proofreading contributes to accuracy during synthesis by DNA polymerase delta. The four-subunit DNA polymerase alpha-primase complex from both HeLa cells and calf thymus is the least accurate replicative polymerase. Fidelity is similar whether the enzyme is assayed immediately after purification or after being stored frozen.(ABSTRACT TRUNCATED AT 250 WORDS)