In this paper, we study the biquadratic optimization problem (BOP) over unite spheres, which is a NP-hard problem. If the objective function is concave, we give a new proof for the equivalence between the BOP and a multilinear optimization problem (MOP). Unfortunately, the objective function may not be concave generally. To overcome this, we present a modified model which is concave, and the optimal solution of the modified model is as same as the original problem. Convergence and numerical results are given to show the performance of our algorithm.