This paper endeavours to show that the major problems in evaluation research are logical rather than methodological, exemplifying the situation by consideration of the radiographic features of rheumatoid arthritis and their value as endpoint measures. The objectives of intervention are only rarely specified with sufficient precision to permit outcome to be assessed in a practical and yet valid manner. This gives rise to difficulty in establishing the nature and validity of the relationship between an attribute, such as erosions seen on radiographs, and the underlying state to be assessed. Relationships can be considered at two levels--the ability of the attribute to act as a surrogate in detection of the underlying state (i.e. at a particular point in time), and the the potential of the surrogate to reveal changes in the underlying state as its course unfolds. Despite a number of serious limitations, radiographic changes emerge as one of the most promising candidates to serve as an outcome measure in rheumatoid arthritis.