It is usually difficult to establish the dynamic model of a launch vehicle that accurately describes its time-varying charac? teristics. Therefore modal identification techniques are particularly necessary to obtain the time-varying dynamic characteristics of launch vehicles under flight conditions. Aiming at the problem of in-flight modal identification of launch vehicles， an output-only re? cursive identification method based on the time-dependent autoregressive moving average model is developed by using exponential? ly weighted mechanisms to track the time-varying characteristics. Without measuring the natural excitation forces， the proposed method can accurately and quickly identify the time-varying modal parameters of launch vehicles by exclusively using the measured response signals. Taking the CERES-1 launch vehicle as an example， time-varying modal parameters before liftoff and during the flight phase are accurately estimated by processing the flight telemetry data. Identification results are consistent with the variation of the finite element analysis results， demonstrating the high achievable accuracy of the proposed method. The proposed in-flight mod? al identification method can obtain the full-cycle modal information of launch vehicles， which meets the engineering requirements for the finite element model updating and attitude control system design.