Abstract: According to sample of methane sensor,the paper put forward a method of correcting nonlinear error of methane sensor,which can identify cont rary model characteristic of methane sensorcorrectly based on the least square support vector machine.It introduced principles of correction method of SVM regression estimation and LS-SVM in details.The method does not make use of any priori knowledge about contrary model function and can ensure that extremal solution is optimal and has generalization ability.The application result showed that the detection precision of methane sensor is 0.4% with the method,the result of nonlinear correction are great satisfaction.