Pillar burst is one of the most frequent dynamic disasters in deep mining, which poses a serious threat to safe and efficient mining. In this study, the failure mechanisms and precursors of pillar burst are investigated by active ultrasonic survey and passive acoustic emission (AE) monitoring in uniaxial compression tests on Zigong red sandstone. Combining active and passive AE monitoring data, a P-wave velocity tomography inversion is performed to analyse the temporal and spatial variations of P-wave velocity structure during the sample failure. Results show that the velocity structure of the sandstone sample is highly heterogeneous during loading, and a low-velocity zone emerges, within which most of the acoustic emission events are present. The dispersion of P-wave velocity reflects the global variations of P-wave velocity. It changes drastically during the peak stage, and increases with the ongoing loading. The AE events differ significantly between the pre-peak and post-peak stages. In the pre-peak stage, AE events are randomly distributed in the sample, while in the post-peak stage, clustered AE events are identified. In addition, it is found that using the homogeneous velocity structure for AE events location results in a higher positioning error. The decreasing b value before the eventual failure of the sample indicates that large-scale crack activities are intensified, leading to the increase of sample heterogeneity, which also proves the necessity of applying the heterogeneous velocity structure for AE events location. The research results can be further used for on-site pillar stability monitoring, and the periodic P-wave velocity tomography provides precursors for pillar bursts.