Abstract: In direction-of-arrival（DOA）estimation using acoustic vector sensor（AVS）arrays，the conventional multiple signal classification（MUSIC）algorithm suffers from noise subspace distortion due to unequal noise power between the pressure and particle velocity channels，leading to spurious peaks in the spatial spectrum. To address this issue，a modified MUSIC（MMUSIC）algorithm based on noise subspace reconstruction is proposed in this paper. By establishing a signal model for AVS arrays，the mechanism by which channel noise mismatch introduces “virtual sources” is analyzed，and then the noise subspace separation criterion is accordingly reformulated. A revised spatial spectrum function is then constructed to suppress pseudo-peaks. Simulation results demonstrate that，under challenging conditions，including low signal-to-noise ratio（SNR），closely spaced sources，and limited snapshots，the proposed MMUSIC algorithm achieves lower root mean square error（RMSE）and higher angular resolution compared to conventional beamforming（CBF），minimum variance distortionless response（MVDR）， and standard MUSIC. The results indicate that the proposed method effectively restores the orthogonality between signal and noise subspaces，mitigates virtual-source interference，and thereby enhances the resolution and robustness of DOA estimation for AVS arrays in complex environments.