In the phase of conceptual design, it is required to obtain an prediction method which is time-saving and accurate in determining the separation boundary. The current prediction methods only refer to the interaction flow field of the carrier at the UAV centroid, resulting in inaccurate attitude prediction. Due to the trajectory-attitude interaction, it is further affected that the accuracy of trajectory prediction. In this paper, the aerodynamic response model based on neural network is firstly constructed, on the basis of studying the calculation method of equivalent angle of attack, angle of side-slip and dynamic pressure, a fast and precise prediction method of UAV separation trajectory and attitude considering double-reference-points is proposed. Compared with the results of CFD-6DOF, for a system about the 300 mm wingspan UAV airdropped from the 4 084 mm wingspan airborne, the deviations of the method are less than 5mm in space position and 0.24° in attitude angle. The prediction accuracy of attitude is improved by an order of magnitude compared with other prediction methods. It takes 3~5 d to calculate a separation trajectory using the numerical computational method, but only about 120 s using the method presented in this paper, which greatly improves computing efficiency. The method is suitable for the fast calculation of the separation trajectory and attitude of small stores with lifting surface dropped from large airborne, and can provide a reference for the separation safety evaluation in the phase of conceptual design for aircraft.