The aim of the study was to establish the benefit of an additional hypothetical laboratory criterion for rheumatoid arthritis (RA), comprising positivity for antikeratin antibody (AKA) and/or antiperinuclear factor (APF). The tests were applied to a series of 308 hospital patients with various recent-onset inflammatory joint diseases who were followed for 3 years. The performance of APF and AKA was compared with rheumatoid factor (RF). The most sensitive (.72) but the least specific (.86) test for RA was the latex test. The most specific (.96) but the least sensitive (.33) test was AKA. Waaler-Rose and APF were intermediate. AKA and/or APF positive patients had significantly more erosions than patients negative for these autoantibodies. Despite the impressive performance characteristics of APF and AKA, the actual classification impact achieved, as compared to using RF as the sole laboratory criterion, turned out to be moderate. This is because the criteria proved to be interrelated. Unlike RF, AKA and APF are not suited to the general laboratory, at least not in their present form. Moreover they so far lack the broad data base of RF.