Damage of bone and cartilage caused by rheumatoid arthritis is visualized on radiographs as erosions and joint space narrowing. The focus of assessment until recently was on progression of damage. The first evidence that drug therapy might influence the course of rheumatoid arthritis appeared more than 30 years ago after the development of a method for scoring these abnormalities. A decade and a half ago additional data became available to validate the term 'disease modifying antirheumatic drugs' when sulphasalazine was shown to slow radiographic progression . Around the turn of the century it became obvious that radiographic progression could be stopped completely by current therapy in a large proportion of patients followed for 1–5 years, and it was appreciated that a number of patients had lower erosion scores in follow-up films. During the same time period scattered reports were appearing of repair of erosions, many with equivocal supporting evidence. Although a few studies presented images that were convincing, no studies have been performed to eliminate reader bias or to examine whether there are specific structural features that permit recognition of repair when the reader is unaware of the true sequence of films. The observation of negative scores, together with credible reports that healing might be real, posed the question of whether negative scores were detecting real structural improvement or artifact and, if real, whether the observations were clinically relevant or were trivial There are additional reasons why these reports of healing and the negative scores drew attention. Of particular importance, therapy with TNF blockers – which are more potent than previously employed therapies – resulted in lower levels of disease activity than were previously seen. As inflammation is a major driver of damage, an absence of inflammation a priori is a prerequisite for halting progression, and for making possible reversal of damage. In addition, data have been presented suggesting that TNF blockers inhibit radiographic progression even in the presence of some persistent inflammation. Presenting radiographic data by probability plots also revealed much more clearly the number of patients with negative scores by presenting the individual data of all patients, and this provided new insight into the potential magnitude of repair . There is currently one trial in which a relatively large number of patients with negative erosion scores supports repair occurring on a group level. Although the appraisal of repair on a group level is a relatively simple statistical matter, translating repair from the group level to an individual patient is not straightforward. The null hypothesis that there is no change from baseline within the group can be rejected if the mean change is below zero and the entire 95% confidence limit is below zero, which occurred in the TEMPO trial. In contrast, a negative change score in an individual patient can be composed of 'true repair', of measurement error or of an image artefact such as rotation hiding a tangential erosion. The interrelationship of these three components is unknown and different in each patient. This argument is also pertinent in evaluation progression scores. In a preliminary study within the OMERACT working group on repair we investigated whether a group of experts would agree on the presence of repair in a set of individual joints. The conclusions were that repair indeed exists according to the majority judgements of the panel, but when the time sequence was blinded it was not possible for expert readers