OBJECTIVE: To investigate the dynamic changes of heart rate variability (HRV) in acute cerebral vascular disease, in order to make a criterion for evaluating cerebral function, and establish an assessment for risk stratification after acute stroke. METHODS: All 35 patients who suffered from cerebral vascular disease in the distribution of the carotid artery system were studied from 1 to 7 days following the stroke. HRV analysis with 24-hour Holter tapes was performed every day or every other day continuously for 5 days. According to GCS, the patients were divided into two groups. Critical cerebral vascular disease (CCVD): a state of severe cerebral dysfunction with the Glasgow coma scale (GCS) value between 3 and 8. Non-critical cerebral vascular disease (nCCVD): a state of slightly cerebral dysfunction with the GCS value between 9 and 15. RESULTS: Of the 35 patients, 17 were classified to CCVD, 18 to nCCVD. The CCVD patients had significantly decreased HRV, especially the standard deviation of all R-R intervals (SDNN) and total frequency (TF). The 24-hour heart rate variability graph lost its circadian variation showing a low and plain curve during the whole day. There was a significant correlation between the decrease of HRV and bad prognosis. The borderline of HRV for prognosis was: SDNN<or=60 ms, TF<or=1 500 ms(2)/Hz, high frequency (HF)<or=600 ms(2)/Hz, low frequency (LF)<or=1000 ms(2)/Hz. Dynamic observation proved that if the components of HRV reached this standard, it always indicated bad outcome. On the contrary, if those components increased continuously and were higher than those of the above, it was always indicative of good outcome. These changes always advanced GCS. Prognosis had a high correlate with TF, LF, HF and GCS. Logistic regression showed that only TF, HF and GCS were the main factors affecting the prognosis. The accurate rate of death predication was 88.89 percent; the accurate rate of survival was 82.14 percent. It was much better than the traditional GCS. CONCLUSION: HRV decreases significantly in CCVD and loses its circadian variation. Dynamic observation of HRV offers a good marker for monitoring cerebral function and the progress of disease. HRV could be considered as an objective, non-traumatic and sensitive indicator of the cerebral function and prognosis.