Ideally HIV antibody tests have to be both extremely sensitive and able to recognize all known HIV subtypes. Three patients whose sera failed to react with a synthetic oligopeptide-based HIV antibody test are described in this report. The patients were a Pakistani male infected recently, an Australian male infected for several years, and a Ugandan woman with AIDS. The presence of anti-HIV antibodies was confirmed by means of a standard algorithm with different assay formats. All three sera failed to react in one antiglobulin enzyme-linked immunosorbent assay (ELISA) (Bioelisa HIV-1+2, Biokit SA). No single underlying reason could be identified for the assay failure in the three cases. The first patient, probably infected recently when first tested, was strongly positive by the same assay a year later, confirming the relative insensitivity of oligopeptide assays reported previously for detecting the early antibody response. The other two patients appear to have been infected for several years. Although unlikely to have been infected with a non-clade B virus, the sample from patient 2 lacked detectable antibody to the transmembrane glycoprotein (gp41), the site of the synthetic oligopeptides. Patient 3, of Ugandan origin, was found to be infected with a non-clade B virus. Although her serum reacted strongly to subtype B gp41 in Western blot, it failed to react in the antiglobulin ELISA. Since there appears to be no single common explanation for these three failures there is little opportunity to identify prospectively those situations where testing using assays employing synthetic oligopeptides on the solid phase is likely to fail.