Abstract The present work assesses the potential of rapid alternating movement analysis for detecting movement disorders like Parkinson's disease. Rapid alternating wrist movements were recorded by a diadochokinesimeter for patients with Parkinson's disease ( n = 10) and healthy controls ( n = 20). An index of irregularity was computed for each individual as the density of jerk singularities (i.e. zero-crossings) during the movements. Several scales of analysis (i.e. “coarseness”) were used for detecting the jerk events and two methods were compared for all of these scales: (1) automatic classification by means of a threshold that optimally separates the indexes of irregularity of the patients from those of the controls, and (2) statistical decision (normal or abnormal) based upon a distribution of indexes of irregularity obtained from a large population of normal subjects. The results showed that (1) two scales of analysis were sufficient and that (2) both methods presented similar performances (e.g. sensitivity = 1.00, specificity = 0.85, efficiency = 0.90). However, statistical decision should be preferred because of its simplicity. The possibility of automatic detection of movement disorders from alternating movements is discussed.