The invention is directed to a method of 3-dimensional (3D) comb filter video decoder. The composite video signal is sampled to obtain multiple sampling data F.sub.mP.sub.x,y, which represents the y.sup.th pixel of the X.sup.th line in the m.sup.th frame, in which m, x, y are positive integer. After using F.sub.m+1P.sub.x,y, F.sub.mP.sub.x,y, F.sub.m-1P.sub.x,y, and F.sub.m-2P.sub.x,y to judge whether the composite video signal is the motion state or the still state. The motion detection is based on the composite video signal without separation of luma and chroma. As a result, the motion detection can be precisely judged.