Trajectory sensitivities provide a way for quantifying the variation of trajectory resulting from（small） changes of parameters and/or initial conditions. The trajectory sensitivity has been applied in power systems， such as for parameter identification， dynamic performance analysis and so on. The calculation of trajectory sensitivity based on the definition itself is usually time-consuming. It relies on the numerical integration， is liable to errors of model and parameters， and cannot provide the physical understanding about the sensitivity. A convolution method for calculating trajectory sensitivity is proposed in this paper， by use of typical system dynamic trajectories， which can be obtained by measurement. Based on the analytical characteristics of the measured trajectory in frequency domain， it is demonstrated that the trajectory sensitivity can be expressed as the linear composition or convolution of the step response and the impulse response. Power system example shows that the proposed method is simple， effective and feasible.