The invention relates to the field of neural networks, in particular to a neural network pruning compression algorithm based on automatic search and knowledge distillation. The method specifically comprises the following steps: step 1, for a convolutional neural network, searching an optimal pruning ratio pl for a neural network l layer; step 2, pruning the layer 1 of the convolutional neural network according to an optimal pruning proportion; step 3, inputting training set samples to the student network and the teacher network, and performing forward propagation; step 4, performing optimization training on the student network by using a knowledge distillation algorithm; and step 5, carrying out back propagation on the student network to update the parameters, optimizing the updated parameters, and then skipping to the step 4. Pruning and precision recovery of the convolutional neural network are realized, and the problem that the convolutional neural network compresses and depends onan original data set is solved.