It is the extremely important to the intelligence traffic that the automation characters recognition of license plate system. The LSSVM has been successfully adopted to make a multi-class classifier on the small samples, nonlinear data. It is presented the LSSVM model methodology on the characters recognition system in the paper. A typical characters recognition of license plate system is composed of the license location, characters extracted and character recognition. The improved genetic algorithm is applied to the optimized selection parameters on the LSSVM model. It put emphasis on the parameter optimization LSSVM model of the characters recognition method of license plate system in the paper. Experiments showed the method is not only feasible, but also have high accuracy, good stability, better computational speed and good generalization ability. The results indicates that the LSSVM strategy improve recognition rate and therefore has more practicability.