Error-prone polymerase chain reactions (epPCRs) are often used to introduce mutations in random mutagenesis, which has been used as a tool in protein engineering. Here, we developed a new method of epPCR using heavy water as a solvent instead of normal water (H(2)O). Rhodopsin cDNA of the Ayu fish (Plecoglossus altivelis) was used as a template and was amplified using five different conditions: (A) 100% H(2)O with no Mn(2+), (B) 100% H(2)O/0.6mM Mn(2+), (C) 99% D(2)O with no Mn(2+), (D) 99% D(2)O/0.6mM Mn(2+) and (E) 99% H(2)(18)O with no Mn(2+). The 13,960 (for each of the conditions A to D) and 33,504 (for condition E) base pairs were sequenced. A maximum error rate of 1.8×10(-3)errors/bp was detected in condition D, without any particular hot-spot mutations. A high preference for AT→GC transitions was observed in condition D, whereas a high preference for transitions over transversions was observed in condition C. All of the mutations observed in condition E were transversions. When conditions A and C were applied to another template, the honeybee actin gene, the results were comparable to those for Ayu rhodopsin. Based on these results, the use of heavy water, instead of H(2)O, as a solvent for epPCR can introduce random mutations without positional bias, template dependency or decreased yield. Our new epPCR method, and possibly combining the use of D(2)O and H(2)(18)O, may be a powerful random mutagenesis technique.