A comparative evaluation of existing assignment criteria for rheumatoid arthritis has been undertaken as a preliminary to exploring how simpler and more efficient standards could be developed. The performances of alternative formulations of individual criteria have been evaluated, and the various criteria have been assessed singly and in combination. The two individual criteria giving rise to the greatest difficulty are morning stiffness and pain and/or tenderness, with its optional formulation. An alternative criterion, pain in at least three sites on squeezing the MCP or MTP joints, performed well. Taken in combination, better epidemiological discrimination was achieved with the threshold for the 1958 ARA criteria midway between 'definite' and 'classical' disease, with six criteria fulfilled. Most of the discriminatory power of these criteria stemmed from inclusion of radiographic and serological characteristics, which is scarcely surprising. The Rome criteria for Inactive RA and the New York criteria for RA, which were derived from them, achieved better discrimination, which suggests ways in which improvements in assignment criteria might be approached. The 1987 revision of the ARA criteria achieves disappointingly little, as important difficulties have not been resolved.