Replication across unrepaired DNA lesions in mammalian cells is effected primarily by specialized, low fidelity DNA polymerases. We studied translesion DNA synthesis (TLS) across a benzo[a]pyrene-guanine (BP-G) adduct, a major mutagenic DNA lesion generated by tobacco smoke. This was done using a quantitative assay that measures TLS indirectly, by measuring the recovery of gapped plasmids transfected into cultured mammalian cells. Analysis of PolK(+/+) mouse embryo fibroblasts (MEFs) showed that TLS across the BP-G adduct occurred with an efficiency of 48 +/- 4%, which is an order of magnitude higher than in Escherichia coli. In PolK(-/-) MEFs, bypass was 16 +/- 1%, suggesting that at least two-thirds of the BP-G adducts in MEFs were bypassed exclusively by polymerase kappa (polkappa). In contrast, poleta was not required for bypass across BP-G in a human XP-V cell line. Analysis of misinsertion specificity across BP-G revealed that bypass was more error-prone in MEFs lacking polkappa. Expression of polkappa from a plasmid introduced into PolK(-/-) MEFs restored both the extent and fidelity of bypass across BP-G. Polkappa was not required for bypass of a synthetic abasic site. In vitro analysis demonstrated efficient bypass across BP-G by both polkappa and poleta, suggesting that the biological role of polkappa in TLS across BP-G is due to regulation of TLS and not due to an exclusive ability to bypass this lesion. These results indicate that BP-G is bypassed in mammalian cells with relatively high efficiency and that polkappa bypasses BP-G in vivo with higher efficiency and higher accuracy than other DNA polymerases.