Abstract: In order to solve the direction-finding resolution problem of multi-sources in the low Signal-to-Noise Ratio(SNR)a virtual ESPRIT algorithm for a pyramid-shaped acoustic vector sensor array is proposed.Based on the array expansion and the Gaussian noise suppression of the fourth-order cumulants,the proposed algorithm calculates the fourth-order cumulants of different output componerits of the acoustic vector sensor,which is equivalent to put a virtual array with the same structure as the original array in the three-dimensional direction.By doing so,one can construct a rotation invariant matrix that includes angular information for direction-finding.By studying the Cramer-Rao Bound of the proposed algorithm,we show its performance is affected by the Signal-to-Noise Ratio(SNR),the number of snapshots and the elevation of the source.The simulation results show that the proposed algorithm has better noise suppression capability and higher resolution Direction-Of-Arrival(DOA)estimation performance than the traditional ESPRIT algorithm with acoustic vector sensor array.