<p><b>Abstract</b>—Clustering of acoustic emission pulses was studied on a mechanically preloaded sandstone sample subjected to high-temperature heating. Mechanical loading was performed uniaxially to a load close to failure, with the appearance of surface signs of large cracks. The sample was then heated to 650°C with recording pulses of thermally stimulated acoustic emission (TSAE). Pulses were clustered based on their similarity established by waveform cross-correlation. Three clusters were identified, each containing approximately ten TSAE pulses and potentially corresponding to a specific source, presumably an individual large crack. The distribution of cluster pulses over time and in amplitude during heating is different for each cluster. This may indicate both different activation time and activation threshold of a corresponding crack as well as different rate of crack growth. Separately, a control thermal stimulation experiment was conducted on a sandstone sample without mechanical preloading. No clusters of TSAE pulses were identified for this sample. Additionally, the parameters of TSAE pulses of the identified clusters were analyzed. It is found that the clusters do not form compact, isolated groups in the parameter space, but are distributed against the background of other pulses. The check of the pulses by their parameters based on the physically substantiated criteria has shown that two of the three clusters included several pulses that were inconsistent with the physical features of the experiment. The hierarchical clustering of all TSAE pulses in the parameter space did not reveal groups that have any similarity with the original clusters. The stability of initial cluster identification in the parameter space was tested using decision tree ensemble classification. The third cluster was recognized with the simplest training criteria. The dynamics of pulses of this cluster is most similar to the growth dynamics of a crack activated at a specific heating temperature.</p>