We demonstrate the application of a biased Monte Carlo method for the optimization of protein sequences. The concept of configurational-biased Monte Carlo has been used, but applied to sequence/composition rather than coordinates. Sequences of two-dimensional lattice proteins were optimized with the new approach and results compared with conventional Monte Carlo and a self-consistent mean-field (SCMF) method. Biased Monte Carlo (MC) was far more efficient than conventional MC, especially on more complex systems and with faster cooling rates. Biased MC did not converge as quickly as SCMF, but often found better sequences.