Freezing in Parkinson's disease is a severe and disabling problem of unknown aetiology. The aim of this study was to analyse the temporal pattern and the magnitude of the electromyographic activity of the lower limb muscles just before freezing and to compare this with a voluntary stop and ongoing gait. We recruited 11 patients with a mean age of 64.8 years (SD 5.1) and a mean Unified Parkinson Disease Rating Scale (part III--off) score of 29 (SD 7.9). Within a standard 3D gait laboratory setting, surface electromyographic (EMG) data of the tibialis anterior (TA) and gastrocnemius (GS) muscles were collected using a portable EMG module. Patients in the off-phase of the medication cycle performed several trials of normal walking and voluntary stops or were exposed to freezing-provoking circumstances. Filtered EMG signals were rectified, smoothed and expressed as a percentage of the gait cycle. EMG onset was determined using a preset threshold, corrected after visual inspection. The magnitude of EMG was calculated by integrating EMG signals (iEMG) over (real) time. To control for the altered timing of activity, iEMG was also normalized for time (iEMGnormt). Analysis of variance of repeated measures analysis showed that significantly abnormal timing occurred in the TA and GS muscles with overall preserved reciprocity. Before freezing, TA swing activity already started prematurely during the pre-swing phase, whereas it was significantly shortened during the actual swing phase. For the GS muscle, a similar pattern of premature activation and termination was found during the stance phase before a freeze. GS activity also showed prolonged bursts of activity during the swing phase, not present during the normal and stop condition. Total iEMG activity of both TA and GS was significantly reduced during the pre-freezing gait cycles. However, when controlling for the altered duration of the bursts, the average iEMGnormt increased, as did the peak EMG in TA. In GS, iEMGnormt was not different in the three conditions. In conclusion, our data show that a consistent pattern of premature timing of TA and GS activity occurred before freezing, which was interpreted as a disturbance of central gait cycle timing. The total amount of EMG activity was reduced in both lower limb muscles due to the shortened time in which the muscles were active. In contrast to GS, activity in TA showed increased amplitudes of the EMG bursts, indicating a compensation strategy of pulling the leg into swing. The observed changes contribute to insufficient forward progression, deceleration and eventually a breakdown of movement.