Background: The Action Research Arm (ARA) test is a performance test of upper extremity motor function which consists of 19 items divided into four hierarchical subtests. This multidimensionality has not yet been tested empirically. Objective: To investigate the dimensionality of the ABA test. Design: Cross-sectional study involving a sample of 63 chronic stroke patients. Methods: A Mokken scale analysis was performed. Results: The Mokken scale analysis revealed one strong unidimensional scale containing all 19 items, of which the scalability coefficient H was 0.79, while H per item ranged from 0.69 to 0.86. The reliability coefficient rho equalled 0.98, indicating a very high internal consistency. A subset of 15 out of 19 items showed an invariant hierarchical item-ordering. Conclusion: The ABA test is a unidimensional scale. The use of subtests, as proposed in the original description of the instrument, is not supported by the present findings. The 15-item scale presented here can be used for adaptive testing, i.e. using only a selected subset of items based on prior knowledge about the patient's abilities, thus minimizing testing time.