As we know most of QCD results show the logarithmic behavior such as the coupling constant $\alpha_s$, moments of $xF_3$, i.e., $M_s$, etc. But in some reactions of deepinelastic scattering (where x is small) the QCD predictions do not fit the experimental data, and shows non-scaling behaviors. Such scaling violation is largely devoted by $\frac{1}{Q^2}$ terms in QCD calculation which we neglected. One of the most important feature of $\frac{1}{Q^2}$ dependence is the highertwist contribution which origined from Wilson operator product-expansion. I will discuss here the important feature of higher twist terms and their origin from QCD.