Antibodies to glutamic acid decarboxylase (GADAb) are found in Stiff-Person syndrome, type 1 diabetes, cerebellar ataxia and other neurological disorders (such as epilepsy and myoclonus) involving the GABAergic ways. GADAb are usually detected by immunohistochemistry (IHC), radioimmunoassay (RIA) or enzyme-linked immunosorbent assay (ELISA). This study analysed the serum of 14 patients with neurological disorders who were positive by IHC for GADAb. The performance of a commercial RIA was compared with in-house immunoblotting and ELISA methods using recombinant GAD65 (rGAD65). RIA was positive in 14 of 14, immunoblotting was positive in seven of 14 and ELISA in 12 of 14. There was no correlation between the RIA result and the ELISA optical densities. Using a sodium thiocyanate chaotrope system with ELISA to determine antibody affinity, we found no significant correlation between antibody affinity and the RIA result. A consensus should be defined concerning which assay could be used as the gold standard for detecting GADAb. The most intriguing finding was that GAD antibodies from uncomplicated diabetics do not appear to recognize GAD in frozen sections from the rat cerebellum, whereas GAD antibodies from neurologically compromised diabetics do. A working proposal is therefore that type 1 diabetic patients with unusual neurological symptoms should be tested for GADAb both by RIA and IHC.