As it has been widely and wrongly believed that noises at any frequency of permanent magnet synchronous motor (PMSM) are positively correlated with subjective annoyance, reducing sound pressure level or sound power level of PMSM is mostly regarded as the only optimisation goal of sound quality (SQ). This study presents a sensitive critical band (SCB) diagnostic method for SQ of PMSM for electric vehicles (EV). First, a new acquisition method of near-field noises without sound attenuation of PMSM was proposed and a neural network model for SQ evaluation was established to determine the specific operating condition with the worst SQ. Second, the band-pass filter of CB (BPFCB) and band-stop filter of CB (BSFCB) were designed to diagnose positive SCBs or negative SCBs in which noises were positively or negatively correlated with the subjective annoyance, respectively. Finally, the major excitation sources of abnormal noises in SCBs were diagnosed by using the analytical calculation. An 8-pole-48-slot PMSM was considered as a case study to testify to the effectiveness and practicability of the proposed method. It may provide a new route for the precise SQ optimisation of PMSM.