The polymerase chain reaction (PCR) is used universally for accurate exponential amplification of DNA. We describe a high error rate at mononucleotide and dinucleotide repeat sequence motifs. Subcloning of PCR products allowed sequence analysis of individual DNA molecules from the product pool and revealed that: (1) monothymidine repeats longer than 11 bp are amplified with decreasing accuracy, (2) repeats generally contract during PCR because of the loss of repeat units, (3) Taq and proofreading polymerase Pfu generate similar errors at mononucleotide and dinucleotide repeats, and (4) unlike the parent PCR product pool, individual clones containing a single repeat length produce no "shadow bands". These data demonstrate that routine PCR amplification alters mononucleotide and dinucleotide repeat lengths. Such sequences are common components of genetic markers, disease genes, and intronic splicing motifs, and the amplification errors described here can be mistaken for polymorphisms or mutations.