Tremor, e.g. in Parkinson patients, often shows large spontaneous fluctuations in severity over the day, to such an extent that a short observation is usually not sufficient to assess the overall severity or the effect of a treatment. Since momentary impressions of the tremor can thus be misleading, long-term ambulatory recordings would be helpful in the evaluation of severity and treatment effectiveness. As existing methods for long-term tremor registration have several shortcomings, a new method is proposed: an algorithm was designed to discriminate tremor from other movements and to describe the amount (i.e. the proportion of tremor or movements per time unit) as well as the intensity (i.e. average acceleration amplitude) of the two types of movement. In the evaluation of the severity of tremor both the amount and intensity of tremor episodes are of importance. The algorithm was tested on 24-h analog tape recordings of wrist-movement in 10 young and 10 aged controls, as well as in 8 patients with tremor--both before and after a tremor relieving thalamotomy. The algorithm scored movements as 'tremor' exclusively in patients prior to the operation. Fluctuations in tremor severity over the day were detected, and tremor could be discriminated from non-pathological movements. Moreover, following thalamotomy, motor slowing (bradykinesia) was detectable using this algorithm. Based on these test results, a miniaturized device in wrist-watch format is now being developed for long-term registrations.