A potentially vast pool of human antibodies with novel specificities for diagnostic and therapeutic purposes can be generated in Escherichia coli. Antibodies to infectious agents have already been isolated by amplifying the heavy and light chain repertoires of donor lymphocytes and they have even been rescued many years after the initial infection from memory cells cultivated in SCID mice. Eventually, however, the creation of extremely large and diverse libraries from the naive antibody repertoire of unactivated B lymphocytes or by gene synthesis using random oligonucleotides for the hypervariable regions could provide a rapid means of obtaining human antibodies to any particular antigen. An important breakthrough for exploiting the potential size and diversity of these libraries has been the development of systems for the surface display of antibodies that are physically linked to their own genes. This allows large numbers of clones to be screened simultaneously and antibodies with affinities of up to 10(8) M-1 have already been obtained using these vectors. It seems quite feasible, therefore, that antibodies with affinities approaching those obtained in the secondary immune response can be obtained by systematically optimizing the strategies for making antibody libraries. Furthermore, it might be possible to establish extremely large antibody repertoires in E. coli by the in vivo recombination of phage and plasmid antibody libraries. The affinity of the selected antibodies could be increased by chain shuffling or random mutagenesis followed by several rounds of selection under increasingly stringent conditions.