Random mutagenesis combined with high-throughput screening is a versatile strategy for improving protein functions or creating artificial enzymes (1,2). Several methods for introducing random mutations in vitro have been reported (3). Among these, error-prone PCR mutagenesis, based on inaccurate copying by DNA polymerase, is the most commonly used technique to introduce random point mutations (4). However, the error-prone PCR method has an inherent drawback of biased occurrence of amino acids as the result of single base replacements in the triplet codons. For example, mutation from AUG (Met) to UGG (Trp) is unlikely to take place. To achieve a non-biased random replacement on the amino acid level, oligonucleotide-directed mutagenesis (5) and cassette mutagenesis (6,7) have been carried out. These methods are limited to a defined region of the gene and can not introduce mutations at random positions. To introduce codon-based mutations at various positions, the split-and-mix method (8) or other synthetic methods of constructing DNAs using dinucleotide or trinucleotide units have been attempted (9–11). These synthetic methods may be applicable only to introduce mutations within a narrow range of the target gene.