To solve the problem of initial attitude estimation for inflight loitering munition in the absence of reference attitude, a fast initial attitude estimation algorithm for adaptive reference vector weight (AFCF) was put forward based on the Attitude and Heading Reference System (AHRS) designed by adopting low-cost magnetic,angular rate and gravity MARG sensor. First of all, a fast error calibration method for three-axes sensor was put forward; Then,the attitude estimation was carried out by adopting the fast complementary filtering algorithm, and the impact of weighting function on initial attitude estimation and convergence was analyzed; subsequently the method for adaptive reference vector weight and adaptive attitude estimation was proposed; finally, high-precision MTI sensor data was used to verify the algorithm,then the algorithm was implemented in the low-cost MRAG AHRS, and performance of the improved algorithm was compared with that of the extended Kalman filter (EKF) algorithm. The experiment results and analysis show that the improved algorithm can achieve a convergence at the initial time when MTI sensor data is used under dynamic conditions, approximately 4s earlier than the fast complement filter (FCF) algorithm; the calculation precision is ±0.6°, and the initial precision is obviously better than FCF.Furthermore, the hardware test indicates that processing time for the improved algorithm is 0.062 ms, accounting for 1/9 of the EKF algorithm, with an approximately calculation precision of ±1.3°, which can meet the requirement of fast convergence, high precision and real-time during the attitude measurement.