Scattering from blood limits the contrast between the vessel wall and the lumen in intravascular ultrasound imaging. This makes it difficult to localize the vessel wall, especially on still images. This paper presents a method for automatic detection of vessel walls and reduction of blood noise based on correlation of the RF-signal between adjacent frames. The ultrasound RF-signal is quadrature demodulated, digitized, stored in memory, and transferred to a computer for processing and analysis. The absolute value of the cross-correlation coefficient between two adjacent frames is used to differentiate between stationary and fluctuating signals. Models and numerical calculations presented in this work indicate that the cross-correlation coefficient obtained from a radially dilating vessel wall will be larger than 0.8 under standard 20 MHz imaging conditions. The corresponding value from blood is less than 0.2 for blood velocities exceeding 0.5 cm s/sup -1/. The blood-noise filter is based on detecting this difference in correlation and displays vessel wall regions with no modifications, while regions detected as blood are rejected. A simplified vessel-wall detector that is suitable for real-time implementation is proposed. The performance of this detector and the blood noise filter are demonstrated by in vitro experiments.