This paper proposes a new practical method for estimating forward rate curves using bond prices available in the market. It is intended to improve the least square estimation method proposed by Carleton and Cooper by imposing additional constraints to guarantee the smoothness of the forward rate curves. The resulting problem is a nonconvex minimization problem, for which we will propose an efficient algorithm for calculating an approximate optimal solution. Computational experiments show that this method can efficiently generate smooth forward rate curves without increasing the residual errors in terms of least square fitting. Also, we will compare this result with an alternative and more efficient constrained least absolute deviation method.