Abstract: This paper proposes a probabilisty neural network classification algorithm based on PLS, the algorithm insteads the PNN input and output with the principal component of PLS, so as to achieve dimensionality reduction, and uses limited pattern neuron alternative sample neurons, thereby greatly simplifies the network, optimizes the structure; finally, the new algorithm is applied into practice, the result shows that the new algorithm with less cost can obtain equivalent classification performance which the traditional PNN has.