Apples’ soluble solids content (SSC) is the typical indicator of internal quality inspection, in order to quickly indicate that, probabilistic neural network (PNN) method is studied for the apples’ SSC and classification nondestructive prediction in this paper. Firstly, red Fuji apples were taken as the research objects and the Near infrared spectral data were collected, which are used to establish the prediction model and classification model. The prediction accuracy of original data is 84.8485%, which is demonstrated by the confusion matrix. For improving the accuracy, multiple scattering correction (MSC) is adopted for deleting the spectral information shifts caused by the change in the external environment. To prove the validity of PNN, simulation was done by comparing the classification accuracy, determination coefficient and confusion matrix of PNN and extreme learning machine (ELM) prediction models, the results show that the accuracy of PNN is 15.91% higher than ELM, and the accuracy of PNN model with MSC treatment is 20.45% higher than the same preprocessing of ELM, indicating that PNN model has better effect on apples’ SSC prediction.