Abstract: In order to study the high-temperature acoustic emission (AE) characteristics of 2.25Cr-1Mo steel, tensile experimental AE monitoring is conducted on the specimens at different temperatures. The tensile process is divided into five stages based on changes in the tensile curves and signal parameter characteristics. The AE signal characteristics of each stage are significantly different, indicating that AE is sensitive to changes in the microstructure of the material. Comparison shows that temperature has the most obvious impact on the AE signal characteristics during the second stage. Correlation analysis reveals that signals generated by different AE sources can also have similar characteristics. As tensile deformation progresses, the main frequency of the AE signals generated by 2.25Cr-1Mo steel converges to a lower frequency band as tensile progresses, and this phenomenon is more prominent at high temperatures. For damage identification, a combination of refined composite multiscale dispersion entropy (RCMDE) and support vector machine (SVM) is introduced. The trained model is used to classify AE signals from different damage modes, achieving an identification accuracy of 91.67%.