Objectives: To improve the cross-correlation method for noninvasive, continuous monitoring of cerebral autoregulation, to evaluate this method in humans with intact and impaired autoregulatory capacity, and to compare it to the cuff deflation test. Design and setting: Prospective study in the intensive care unit of a university hospital. Patients and participants: Fourteen patients with severe head injury, six patients with subarachnoid hemorrhage, and nine healthy volunteers. Interventions and measurements: Middle cerebral artery flow velocities and arterial blood pressure were monitored continuously. Aaslid's thigh cuff tests were performed and results were scored using Tiecks' model for autoregulation index. Data were then collected without any patient manipulation. The mean time delay between slow spontaneous oscillations of blood pressure and middle cerebral artery flow velocity was calculated by cross-correlation analysis. Data are expressed as median (lower/upper quartile). Results: Healthy subjects had a higher autoregulation index than patients, 5.0 (5.0/5.5) vs. 3.3 (2.0/4.5). Slow oscillations of blood pressure and middle cerebral artery flow velocity showed a time delay of -2.0 s (-2.7/-1.7) in healthy subjects but were almost synchronal in patients, -0.07 s (-0.5/0.45). Inter-method agreement in diagnosing an intact or impaired cerebral autoregulation was obtained in 108 of 147 examinations of autoregulation (73.5%) and was considered moderate. Conclusions: Cross-correlation analysis may serve as a simple, noninvasive, and continuous measure of cerebral autoregulation. The time delay of -2.0 s in healthy subjects is in good agreement with other studies. Short-term autoregulation tests and monitoring techniques based on slow spontaneous oscillations should not be used interchangeably.