Mismatch repair (MMR) of replication errors requires DNA ends that can direct repair to the newly synthesized strand containing the error. For all but those organisms that use adenine methylation to generate nicks, the source of these ends in vivo is unknown. One possibility is that MMR may have a "special relation to the replication complex" [Wagner R, Jr., Meselson M (1976) Proc Natl Acad Sci USA 73:4135-4139], perhaps one that allows 5' or 3' DNA ends associated with replication to act as strand discrimination signals. Here we examine this hypothesis, based on the logic that errors made by yeast DNA polymerase α (Pol α), which initiates Okazaki fragments during lagging-strand replication, will always be closer to a 5' end than will be more internal errors generated by DNA polymerase δ (Pol δ), which takes over for Pol α to complete lagging-strand replication. When we compared MMR efficiency for errors made by variant forms of these two polymerases, Msh2-dependent repair efficiencies for mismatches made by Pol α were consistently higher than for those same mismatches when made by Pol δ. Thus, one special relationship between MMR and replication is that MMR is more efficient for the least accurate of the major replicative polymerases, exonuclease-deficient Pol α. This observation is consistent with the close proximity and possible use of 5' ends of Okazaki fragments for strand discrimination, which could increase the probability of Msh2-dependent MMR by 5' excision, by a Msh2-dependent strand displacement mechanism, or both.