Korla pear, a native fruit, is famous for its crispy and sweet taste. There is a large difference in the internal quality of pears, due to the changes in soil, water, and light intensity of the ever-increasing planting area for Korla pear. At present, only firmness and soluble solid content are used as indicators for the internal quality evaluation in the pear grading standard. Nevertheless, the indicator “crispness” is not equivalent to the firmness of the fruit. The crispness reflects the tactile and auditory comprehensive perception of force and sound behavior generated in the process of chewing pear flesh. Since it is difficult to measure and explain through clear semantics, the crispness has not been taken as the standard of internal quality evaluation and classification. A sensory testing is widely accepted to evaluate the crispness, providing by experts or trained panelists, but sensory testers are prone to fatigue and low efficiency. Therefore, it is necessary to investigate an approach to accurately detect the crispness of pears to the taste of consumers. In this study, a total of 250 pears were stored at (26±2) ℃ and 20% relative humidity (RH), where the storage time was 0, 10, 20, 30 and 40 d. A texture analyzer combined with an acoustic envelope detector was used to simultaneously collect the signals of force and sound, where 4-6 cylindrical samples were tested in each pear. 15 mechanical parameters and 6 acoustic parameters were extracted from force and sound signals using the peak, particularly on the parameters autocorrelation. The results showed that there was a highly strong correlation with relatively little redundancy in 5 pairs of parameters, including the acoustic power and sound linear distance, the average level of sound pressure and sound peak number, Young ' s modulus and low strain stiffness, average force and work, force difference and force ratio. All the mechanical and acoustic parameters can be directly used to construct the classification model without dimensionality reduction. The artificial neural network (ANN) and support vector machine (SVM) were used to classify the crispness of Korla pears. Three types of parameter datasets were fed to train the ANN and SVM models: mechanical and acoustic parameters, as well as the combination of mechanical and acoustic parameters. A comparison of ANN models showed that the model using a three-layer hidden structure (14 nodes in each layer) achieved the highest classification accuracy. In the SVM with different kernel functions, the model with the quadratic kernel function displayed the best classification performance. A combination of mechanical and acoustic parameters was more applicable to detect the crispness of pear flesh than only mechanical or acoustic parameters. In learning curve, the classification accuracy of the SVM and ANN models achieved 96.1% and 93.8%, respectively. Therefore, the models can meet the requirements of accurate classification for pears with different crispness. This finding can provide practical guidance to evaluate the crispness of pear flesh during harvest, processing, and storage.