Artificial neural networks have been successfully used in classification, formulation optimization, defect diagnosis and performance prediction in ceramic industry. However, an artificial neural network based on the traditional backpropagation (BP) algorithm showed some disadvantages in mapping the nonlinear relationship between the composition and contents of the ceramic materials and their properties. In this paper, a new PSO-Grain (Particle Swarm Optimization Gain) BP algorithm was introduced, and an improved artificial neural network model was employed to predict the properties of an alumina green body. The training performance of the neural network using the PSO-Gain BP algorithm was analyzed and it was indicated the POS-Gain BP based neural network could reduce convergence to local minima and was more efficient than the traditional BP based network. The prediction accuracy of the properties such as linear shrinkage and bending strength using the PSO-Gain BP based neural network was higher than that of the BP based neural network.