OBJECTIVE: To investigate the construct validity and unidimensionality of the Rivermead Motor Assessment, arm section, using Rasch analysis which transforms the ordinal scale into an interval scale with hierarchical ordering of items. SETTING: Twelve stroke units and rehabilitation centres across Belgium, Germany and Switzerland. SUBJECTS: Two hundred fifty-two stroke patients, average (SD) age of 63.5 (13.22) years. MAIN MEASURES: Principal component analysis on the Rasch data gave additional information on unidimensionality. Differential item functioning established the stability of items for gender, age, country, hemiplegic side, stroke type, and time since stroke. RESULTS : Four items did not meet the requirements for the Rasch model and were removed. These were 'Lying, protract shoulder with arm in elevation', 'Pick up tennis ball from table, release on midthigh, return to table, release', 'Same exercise with pencil' and 'Pick up piece of paper from table. Release five times'. Principal component analysis of the remaining 11 items revealed two dimensions in the scale, 'static and dynamic control of movement', and 'coordination'. All items were stable among the subgroups of the sample. CONCLUSION: The Rivermead Motor Assessment, revised arm section meets the required criteria for validity and unidimensionality. It consists of 11 individual items, which yield two subtotals of overall upper limb ability of the patients. It is a valid tool to evaluate upper limb function after stroke.