The use of a multimicrophone acoustic intensity probe is investigated for the measurement of the acoustic velocity power spectrum. The conventional formulation [J. Acoust. Soc. Am.73, 1047–1053 (1983)] for this measurement is found to be sensitive to measurement errors. Three alternative formulations of the velocity estimate are developed that are less sensitive to random error. The various formulations are implemented into a four‐microphone probe for measurement of the three velocity components. Experiments in the nearfield of a piston and a periodically stiffened plate verify the predicted relative advantages of the various estimators.