Although bispecific antibodies (bAb) are widely recognised to offer great promise in the therapy of malignancies and other diseases, significant difficulties in their production and purification have limited their progress to clinical application. Techniques employed to overcome this bottleneck in the production of clinical-grade bAb (starting from traditional bAb preparation by chemical crosslinking or hybrid-hybridoma technology) include improved purification methods for hybrid hybridomas (Allard et al. 1993), selective crosslinking chemistries for bispecific Fab2 (Brennan et al. 1985; Glennie et al. 1987; Carter et al. 1992) and the use of amphipathic helices to drive association (Kostelny et al. 1992). Recently protein engineering has been used to design new formats for bAb. These adesignero bAb are based on antibody Fv or scFv fragments as building blocks rather than whole antibodies (reviewed by Holliger and Winter 1993) and can be expressed in recombinant form as the main product and purified to homogeneity using a one-step purification procedure. One such bispecific antibody fragment is the diabody (Holliger et al. 1993). Instead of a single polypeptide chain with four domains, diabodies are dimers, each chain comprising two domains. Each chain consists of a VH domain connected to a VL domain using a linker is too short to allow pairing between domains on the same chain, thus driving the pairing between complementary domains on different chains to recreate the two antigen-binding sites. Thus, for a bispecific diabody (made from antibodies A and B), the first chain is assembled from the VH domain of antibody A and the VL domain of antibody B and vice versa to create the two chains VHA-VLB, VHB-VLA. Each chain alone is incapable of binding to antigen, but associates with the other chain to form a bispecific diabody (Fig. 1). In diabodies the antigen-binding sites are at opposite ends of the molecule, as shown by X-ray crystallography (Peresics et al. 1994). In contrast to IgG or Fab92 (and presumably bscFv), in which the two binding sites can take up a range of orientations and spacings, the diabody structure is more rigid and compact, with the two binding sites separated by 6.5 nm (less than half the distance in IgG). Molecular modelling indicates that the two Fv heads of the diabody can adopt a range of different conformations and move within 30° of the mean position (Holliger et al. 1996), so their flexibility is clearly restrained. Despite their limited flexibility, diabodies appear to be as effective at recruiting cytotoxic T cells as are bispecific bscFv and Fab92: in the BCL-1 lymphoma model a diabody directed against the idiotope and mouse CD3 was able to activate naive T cells and induce antigen-dependent cytotoxicity towards lymphoma cells at concentrations of a few nanograms per millilitre (Holliger et al. 1996). The potency of the diabody (weight/volume) in the T cell activation and the cytotoxicity assay appears to be similar to that of a bscFv (refolded from inclusion bodies) comprising the same antibody V genes B1 and 2C11 (De Jonge et al. 1995). Likewise when Zhu et al. 1996) compared the effectiveness of a diabody directed against p185 (HER2) and human CD3 with the analogous Fab92 bsAb in inducing cytotoxic T lymphocyte cytotoxicity towards HER2+ tumour cells, they found the diabody to be equally potent. Apart from cytotoxic T cells, other immune effector functions can be recruited using bispecific diabodies, e. g. the complement cascade by retargeting complement component Clq (Konterman et al. in press) on the whole gamut of antibody effector functions (including complement) by Work presented at the 5th World Conference on Bispecific Antibodies under the Auspices of EFIC; 25±28 June 1997, Volandam, The Netherlands