The invention discloses an automatic classification method for ice crystal pictures. The method comprises the following steps: firstly, segmenting a large picture containing ice crystal particles at multiple moments into ice crystal particle pictures at a single moment; determining the number of ice crystal categories according to the observed ice crystal picture data, and establishing an ice crystal particle picture database; dividing the ice crystal particle database into a training set and a test set, training and testing the ice crystal particle database by using a convolutional neural network (CNN) model, adjusting the CNN model, and storing the model with the optimal test effect; and automatically classifying the ice crystal pictures by using the stored model. By means of the method,the observed ice crystal particle segmentation problem can be solved, the problems that a traditional method is low in ice crystal classification accuracy, time-consuming and subjective can be solved, and segmented ice crystal particle pictures can be automatically classified.