Abstract: Support vector regression machine (SVRM) is integrated with the statistics learning theory (SLT) to map training samples into a high dimension space. But sometimes the operation speed and the accuracy of the standard support vector regression machine is not ideal. For a case of linear indivisibility, two relaxation items are added into the objective function of the support vector regression machine in order to reduce two constraint conditions. The weights can be easily adjust ed according to practical requirements by adding two weighting factors. The method is named a new weighed support vector regression machine (WSVRM) for function approximation. The experimental results show that the proposed new type of weighted support vector regression machine has good function estimation and data forecasting capabilities.