Aiming at the fact that parameters in support vector machine(SVM) model were difficult to be identified, a genetic algorithm SVM(GASVM) was proposed to avoid the blindness in parameter choosing and improve the estimation ability of SVM, in which the parameters in SVM and kernel function were searched by genetic algorithm. And it was then applied to the classification for the swell and shrink grade of expansive soils. Five indexes including liquid limit, total swellshrink ratio, plasticity index, water contents and free expansive ratio were adopted as discriminated factors. And the four grades of the expansive soils were the outputs correspondingly. Classification function was obtained through training a large set of expansive samples. And it was shown that the classification method of GASVM was effective and with high accuracy.