RT-PCR amplification of P450 2C6 from rat liver, using primers in opposite orientations of exon 6, resulted in PCR products containing segments of exons joined at non-consensus splice sites. Moreover, many of the PCR products identified were composed of not only a single region containing exonic segments joined at non-consensus splice sites but, instead, of several repeats of the non-canonically joined region. To investigate whether these PCR products represent pre-existing molecules or are generated during the amplification process, the liver cDNA template was replaced by a plasmid containing the P450 2C6 cDNA. Surprisingly, PCR products containing repeats of non-canonically joined exonic segments were again revealed. In some cases the position of this non-canonical joining was a sequence of one or two identical nucleotides; however, there were also a number of products lacking any nucleotide identity at the position of joining. DNA nicking and/or DNA damage is thought to favour recombination during PCR, probably by misalignment of incomplete DNA strands; however, the presence of multiple repeats of the recombined region in the PCR products identified suggests a certain repetitiveness of the underlying mechanism. It is therefore proposed that these products result from a template switching event that occurs several times during a single polymerization step, following a rolling circle model of DNA synthesis.