In response to the challenge of identifying the aircraft body model in closed-loop flight tests, this paper proposes a method based on multi-model theory. A Kalman filter is used to estimate one-step prediction errors of multi-model, then calculates model matching probabilities based on a multivariate Gaussian distribution to identify the most accurate model. Using a small unmanned aerial vehicle (UAV) as a platform, experiments were conducted for test design, closed-loop identification, and multi-model validation. The identified model was used to design the roll control law and conduct flight tests. Results show that this approach can enhance the accuracy of closed-loop identification model, thereby improving control performance.