We introduce a new tensor-based model reduction method for fluid simulation. Our model reduction algorithm accelerates simulation without sacrificing simulation quality. The algorithm reduces calculation time by solving the projection step in the reduced space as opposed to a full grid space. The tensor model reduction extracts a number of basis from a velocity field, using HOSVD(High Order Singular Value Decomposition). The heavily reduced model still maintains important visual features. The reduction and reconstruction process work on the grid based fluid simulation, so our method can be easily incorporated into a typical fluid simulator that solves Navier-Stokes equation.