Abstract: The product distributions of catalytic cracking have a complex functional correlation with feedstock compositions and reaction conditions. The experimental results of three heavy oil samples having various compositions and testing under different reaction conditions were normalized as training data, by using support vector regression (SVR) method, a yield model for gasoline and diesel products of heavy oil catalytic cracking was established. For catalytic cracking of recycle stock, gasoline and diesel yields under different operation conditions were calculated by the generalization ability of SVR model. The optimal operation conditions for maximizing light oil (gasoline and diesel) yield were found by particle swarm optimization (PSO) algorithm. Calculated results show that this model has good fitting effect with experimental data under various reaction conditions. The simulated three-dimensional graphs can effectively illustrate the relationships between product yields and reaction conditions. The optimal reaction conditions obtained by PSO algorithm are as follows: reaction temperature of 530℃, catalyst to oil mass ratio of 7.5 and space velocity of 8 h-1. Under such conditions, the simulated and experimental light oil yields are 42.3% and 41.8%, respectively, the relative error is 1.20%.