In this paper a new controller design, which we Shall call the "trajectory sensitivity optimization" method, is presented to improve the robustness for parameter variations. The method uses the sensitivity trajectory to model the parameter uncertainty and introduces a special quadratic cost function involving an input and output sensitivity term. Necessary conditions are derived to obtain the dynamic controller. The necessary conditions consist of two Lyapunov equations and two controller gain equations that have no closed-form solution. Therefore, a special iterative algorithm was developed to obtain the numerical solution. The method can deal with a wider class of parameter uncertainty than existing methods. Numerical examples show that the method is effective in improving the robustness to parameter variations.