Abstract: Based on standard support vector machines(SVMs), the bound of both the support vector number(and rate) and boundary support vector number(and rate)is proposed and proved.Then the bounds are extended to positive class and negative class respectively.On the basis of the bounds,it is proved that the positive class yields poorer classification and predictive accuracy than the negative class does.Simulation results of both artificial data sets and benchmark data sets show that the conclusion and method in this paper is true and effective.