BACKGROUND/AIMS: The vast number of existing dedicated techniques proves that skin blood flow estimation is an unsolved problem. Specificities of cutaneous vascularization (very low blood velocity, noisy environment, complexity of the vascularization architecture) result in the unsuitability of conventional ultrasonic Doppler techniques (long acquisition time, low spatial resolution). The object here was to present a high-frequency time-domain correlation METHOD: In particular, the difficulties of adaptating this type of measurement (data processing, hardware problem) are pointed out. METHODS: Radio-frequency (RF) backscattered signals, obtained with a modified version of a home-made 20 MHz skin imaging system, are studied. Time shifts between successive windowed sections of the RF signals are determined by the mean of the cross-correlation algorithm. A realignment procedure (to remove the artefacts caused by the movements of the patient and the manipulator) and a stationary echo cancelling procedure (to remove the signals coming from the cutaneous tissues and to permit the detection of very small vessels) are used. RESULTS: In vitro results show that velocity measurements as low as 0.1 mm/s are attainable with a 80 &mgr;m axial resolution, and blood vessels of 100 &mgr;m are detectable. Our technique has also been validated by means of in vivo experiment on an erysipelas located on a human leg. In this way, a 180-&mgr;m-diameter blood vessel has been detected on a M-mode RF image and the corresponding velocity profile has been obtained. CONCLUSION: Further improvements can be expected, and the level of performance obtained in vitro in this work should be also attainable in vivo and should then provide an effective tool for skin physiology and pathology.