A study was designed to evaluate observer variation in the assessment of radiographic deterioration of individual patients using the Larsen grading system. Radiographs of hands and feet of 52 patients were assessed by three observers. Each patient had paired films taken one year apart which were assessed together for change in score. To assess within-observer variation each set of films was read twice by all observers. The average progression was 11.6 (SD 9.0). Analysis of the source of variation showed the single observer replication SD to be 3.7 but that for different observers to be 5.5. This may be interpreted as indicating that to achieve 95% confidence of detecting a true change an increase in Larsen score of 8 is required if the same observer assesses or up to 11 if a different observer assesses.