Abstract: Matched-Field Inversion(MFI)is an effective method which can infer the seabed parameters quickly.But it can only give an averaged estimation in the range-dependent environments,and has limited ability to obtain the bottom density and attenuation in a more complex environment.MFI has been performed usually using a Vertical Line Array (VLA)of sensors which has poor system mobility.The use of a Horizontal Line Array(HLA)can offer an inversion method in moving way.This paper combined components of Matched-field inversion and Reflection-loss inversion to develop a multi-step Bayesian inversion for geoacoustic parameters.This method applies the Posterior Probability Density(PPD)or inversion results from one inversion as prior information for subsequent inversion,the sensitive and insensitive parameters are determined respectively,and the search bounds are gradually narrowed.The Bayesian method and Markov-chain Monte Carlo approach were developed to get the distribution of PPD.In order to verify the inversion method,the statistical properties of transmission loss based on the posterior probability were introduced.The PPD results indicated that the multi-step inversion method was performed better than direct matched-field inversion,and some parameters' inversion results were improved significantly.From the transmission loss distribution,it can be seen that the predicted acoustic fields had smaller errors.