Brainstem auditory evoked potentials (BAEPs) have been widely monitored to prevent hearing loss (HL) during microvascular decompression (MVD) for hemifacial spasm (HFS); however, their predictive value is still unclear. The aim of this study is to investigate the predictive values of the maximum changes in BAEPs and define the best warning indicator and a cutoff value (CV) during HFS-MVD. The clinical data of 93 HFS-MVD patients were retrospectively analysed. The maximum change rates of the latency and amplitude of waves I, III, and V and the interpeak latencies (IPLs) I-III, I-V, and III-V, when BAEPs change most during MVD, were defined. Pure tone audiometry was performed to evaluate hearing loss (HL). Logistic regression, propensity score, receiver operating curve (ROC), and area under the curve (AUC) were used to identify the predictive value of relevant indexes and to determine the CV (with the largest Youden index) of the best index at different levels of HL. The AUCs of BAEPs for predicting HL were 0.98, 0.92, and 0.84 for 50 dB, 30 dB, and 10 dB, respectively. The amplitude of wave V (AwV) was the best single predictive index at all three HL levels. The CV of AwV was 55% (50 dB), 46% (30 dB), and 34% (10 dB). At 50 dB HL, the predictive value of IPLs I-V (AUC 0.89 with CV 0.6 ms) was better than that of LwV (AUC 0.82 with CV 1 ms). BAEPs can predict HL well. AwV is the best single predictive index of all BAEPs. The reduction of AwV by 34% (watching), 46% (reporting), and 55% (warning) can be used as a sliding-scale warning sign. In addition, IPLs I-V (> 0.6 ms) and LwV (> 1 ms) should also be observed and reported during MVD.