A retrospective study was performed to investigate seasonal variation in stroke incidence and to evaluate the hypothesis that cold might be a risk factor. Data were obtained from the central registry of the Hospital de S. João, Porto, Portugal, concerning 4048 patients consecutively admitted for cerebrovascular disease during a period of 33 months. Monthly admissions for stroke and its subtypes were related to mean values of ambient temperature using linear correlation. There was a strong inverse correlation between average temperature and total admissions for cerebrovascular disease (r = -0.72, P < 0.00005), intracerebral haemorrhage (r = -0.66, P < 0.00005), ischaemic stroke (r = -0.46, P = 0.007) and transient ischaemic attack (r = -0.41, P = 0.017). These correlations were independent of any seasonal variation in the number of hospital admissions due to all causes. No relation was found between temperature and subarachnoid haemorrhage. The rhythmometric analysis showed the presence of a statistically significant rhythm with an acrophase in the coldest months. These results support the hypothesis of stroke being a chronorisk disease to which cold might represent a triggering factor.