Criteria for ankylosing spondylitis are useful for two main purposes. First, to provide unity of diagnosis in population surveys; secondly, in the field of therapeutic assessment. This paper is concerned with criteria used largely in the former role, i.e. as an epidemiological tool. Critical formulation and subsequent evaluation of criteria is important if a realistic measure of disease prevalence is to be obtained. The New York criteria have provided the general currency for diagnosis epidemiologically in recent years, but their value, in their originally published form, has been questioned for various reasons. Attempts have been made to improve these criteria or to suggest alternatives. Although increased sophistication of the original New York criteria has doubtless been achieved, the position regarding the diagnosis of ankylosing spondylitis continues to represent a challenge. The main concern is that the present "index of truth" for the disease, radiological sacroiliitis, is all too often a delayed feature, presenting only after pain has been present for some time. A second problem is the difficulty in diagnosing with confidence the early changes of sacroiliac sclerosis and erosion, and the degree of intra- and inter-observer error arising from these changes. Newer techniques such as radioisotope scintigraphy and computerised tomography (CT) are unlikely to replace the traditional pelvic radiograph. The way forward is more likely to be in the direction of how existing data are processed rather than what is used to obtain them. In this vein, suitable mathematical/biological models coupled with computer technology could provide realistic tools through a system of probability weighting of existing criteria.