Evaluating joint involvement in rheumatoid arthritis in a key clinical assessment. We investigated the extent of variation in measurement of joint swelling and tenderness and evaluated the impact of training to standardise methods. Eight observers (medical and nursing staff) examined eight rheumatoid patients for joint swelling and tenderness before and after training in clinical methods. The EULAR handbook for joint evaluation was used for training and assessments were based on the 28 joint count. There was extensive variability in both numbers of swollen and tender joints. Coefficients of variation for articular indices recorded by the 8 observers in individual patients were often high (up to a maximum of 204%), indicating considerable differences between observers. Training had an impact on the assessment of the numbers of swollen joints which increased by a mean of 32% (P < 0.05) and the number of tender joints which increased by 41% (p < 0.01). Training had only a limited impact on the variation among observers in determining the number of swollen and number of tender joints. After training, the mean coefficients of variation were still 59% for swollen joints and 65% for tender joints. These results highlight the extent of variation in clinical assessment of rheumatoid arthritis and show the advantages of training. It leads to increased sensitivity of measurement. Standardisation appears essential for clinical studies.