Objectives. – To evaluate observer agreement using the Larsen system (LS) and a Modified Larsen system (ML) when assessing individual joints of the hands and wrists in rheumatoid arthritis, and to compare the two systems. To determine the minimally important difference (MID) for the ML. Methods. – hirty radiographs of hands and wrists from 10 patients who presented with RA were graded by two blinded observers, using the LS and then the ML. Patients were followed for a mean of 7.2 years (range: 4–10 years). Inter- and intra-observer agreement were calculated using the kappa statistic with linear incremental weights. Inter-observer agreement was also computed for the summed score, using an intraclass correlation coefficient. Inter-observer error was estimated by calculating the mean and standard deviation of the grading differences between the two observers. Prevalence of damage was calculated as a ratio of damage: no damage and expressed as a percentage. Pairs of radiographs were comparatively graded using a seven-point Likert scale. Results. – The kappa statistic for inter-observer agreement was 0.38 (marginal reproducibility) for the LS and 0.52 (good reproducibility) for the ML (P = 0.004). Using a difference of one grade as perfect agreement, it was 0.56 (good reproducibility) for the LS and 0.87 (excellent reproducibility) for the ML (P = 0.001). Intra-observer agreement was high in both systems. The distribution of ML-grade differences varied according to the level of the Likert scale: for “a little bit worse”, representing the smallest amount of detectable damage progression, the distribution differences peaked around two grades. This value represented a MID 87% of the time. Conclusions. – The LS lacks precision for individual joints. The ML, it is proposed, has more detailed definitions of grades, and is more reliable. When pairs of radiographs were compared, a two-grade difference on the ML was the MID.