Compositions, methods, and kits are provided for efficiently generating and screening a library of highly diverse protein complexes for their ability to bind to other proteins or oligonucleotide sequences. In one aspect of the invention, a library of expression vectors is provided for expressing the library of protein complexes. The library comprises a first nucleotide sequence encoding a first polypeptide subunit; and a second nucleotide sequence encoding a second polypeptide subunit. The first and second nucleotide sequences each independently vary within the library of expression vectors. In addition, the first and second polypeptide subunit are expressed as separate proteins which self-assemble to form a protein complex, such as a double-chain antibody fragment (dcFv or Fab) and a fully assembled antibody, in cells into which the library of expression vectors are introduced. The library of expression vectors can be efficiently generated in yeast cells through homologous recombination; and the encoded proteins complexes with high binding affinity to their target molecule can be selected by high throughput screening in vivo or in vitro.