支持向量机和BP神经网络虽然都可以用来做非线性回归，但它们所基于的理论基础不同，回归的机理也不相同。支持向量机基于结构风险最小化理论，普遍认为其泛化能力要比神经网络的强。为了验证这种观点，本文编写了支持向量机非线性回归的通用Matlab程序和基于神经网络工具箱的BP神经网络仿真模块，仿真结果证实，支持向量机做非线性回归不仅泛化能力强于BP网络，而且能避免神经网络的固有缺陷——训练结果不稳定。-SVM and BP neural networks, although non-linear regression can be used to do, but they are based on different theoretical basis, the return mechanism is not the same. SVM based on structural risk minimization theory, generally considered the generalization ability of neural networks than strong. To test this view, a support vector machine of this writing the general non-linear regression procedures and based on Matlab neural network toolbox of the BP neural network simulation module, the simulation results confirm that support vector machines do not only the generalization ability of non-linear regression in BP network, and neural networks to avoid the inherent shortcomings- the training results unstable.