Time of arrival ( TOA) is a key feature utilized in localizing impacts, with the ongoing advancements in signal processing technology, a variety of time-frequency models can be found in the literature for its measurement, but the selection usually depends on human experience. In addition, considering the systematic error of the measurement model and the uncertainty caused by measurement noise, the traditional TOA-based impact localization method is insufficient. Therefore, an impact localization method for composite structures based on Bayesian estimation and data fusion is proposed in this paper. First, a variety of different time-frequency models are applied to obtain the TOA data of the impact response signal. Then, according to the uncertainty of measurement error of TOA data, the posterior probability density function of impact position is constructed by using the Bayes theorem. Afterward, the posterior distribution of the impact position parameters is estimated using the Markov chain Monte Carlo (MCMC) sampling method, and the normal distribution is used to fit the posterior distribution. Finally, the impact position probability distribution obtained from different TOA data is fused, and the final decision is made by using the fused probability distribution. The feasibility of the proposed method is verified by the drop weight impact experiment on the composite stiffened plate, the average localization error is only 0. 94 cm, which is more reliable and accurate than the traditional TOA-based impact localization method, and also has advantages in robustness and localization time.