A study comparing actigraphy and polysomnography in the detection of leg movements was performed in a group of 35 patients with sleep disorders. Visual scoring used epochs of 5-second duration, and in each epoch, electromyographic activity of tibialis anterior muscle was classified in eight types on the basis of its duration and amplitude. Activity levels of the actigraphic data were logged in 5-second intervals and stored in memory for computer retrieval. To assess the reproducibility of actigraphic detection scoring, in 10 patients actigraphic and polysomnographic data were compared during two consecutive nights. Despite a high correlation between polysomnography and actigraphy in the detection of muscular activity (r = 0.78; P = 0.001), no agreement was obtained between the two analyses according to the Bland and Altman method of concordance (mean bias-117.5). The greater agreement was found for electromyographic activity lasting more than 3 seconds and with an amplitude of more than 50 microV. In the group of patients examined during two consecutive nights the scoring error did not differ significantly. The results demonstrated that (1) despite a significant correlation between actigraphic and polysomnographic data, actigraphy substantially underestimated leg electromyographic activity and therefore could not be used routinely for diagnostic purposes; and (2) because error scoring did not undergo significant changes across the two nights, actigraphy may be used for follow-up evaluation.