AIMS--To compare the predictive value of immunoblotting and enzyme linked immunosorbent assay (ELISA) in diagnosing Lyme borreliosis. METHODS--An ELISA using a whole cell sonicate of the B31 strain of Borrelia burgdorferi was used to screen samples submitted for Lyme borreliosis serology. A total of 1222 serum samples reactive in the ELISA were tested by immunoblotting also using the B31 strain. Patients with other spirochaetal diseases were tested by both methods to assess specificity, while those with erythema migrans were used to evaluate sensitivity. Subjects with different clinical conditions, which may have been associated with Lyme borreliosis, were tested using both techniques. RESULTS--Only 16.3% of serum samples from patients submitted for Lyme borreliosis serology which were reactive by ELISA were confirmed as positive by immunoblotting. This is unlikely to represent a sensitivity problem as 51% of samples from 53 patients with erythema migrans were detected by immunoblotting compared with only 28% by ELISA. Patients whose samples were negative by ELISA were also negative by immunoblotting. Serum samples from patients with relapsing fever were reactive in both ELISA and by immunoblotting, but for other test groups immunoblotting offered increased specificity. CONCLUSIONS--Not all ELISA results could be confirmed by immunoblotting. Yet immunoblotting was both more sensitive and specific than ELISA techniques. As a result of these observations all ELISA results should be serologically confirmed by immunoblotting. Though immunoblotting is not suited to large scale screening of samples, it can be used satisfactorily in conjunction with ELISA methods to improve the predictive value of serological tests for Lyme borreliosis.