There is a large and increasing number of therapeutic proteins approved for clinical use and many more undergoing preclinical studies and clinical trials in humans. Most of them are human or 'humanized' recombinant molecules. Virtually all therapeutic proteins elicit some level of antibody response, which in some cases, can lead to potentially serious side effects. Therefore, immunogenicity of therapeutic proteins is a concern for clinicians, manufacturers and regulatory agencies. In order to assess immunogenicity of these molecules, appropriate detection, quantitation and characterization of antibody responses are necessary. Immune responses to therapeutic proteins in conventional animal models has not been, except in rare cases, predictive of the response in humans. In recent years there has been a considerable progress in development of computational methods for prediction of epitopes in protein molecules that have the potential to induce an immune response in a recipient. Initial attempts to apply such tools in early development of therapeutic proteins have already been reported. It is expected that computer driven prediction followed by in vitro and/or in vivo testing of any potentially immunogenic epitopes will help in avoiding, or at least minimizing, immune responses to therapeutic proteins.