A linearization approach to acoustic inversion is proposed employing distinct ray paths (direct arrival, first surface bounce, and first bottom bounce) for source localization and bathymetry and sound speed estimation. The ray path arrivals are selected from broadband, shallow water, synthetic data using a Bayesian time delay estimation scheme calculating posterior probability density functions of the delays in an efficient way. A linear system is then formed relating unknown parameters and arrival time data. The regularization method is used for the solution of the linear system [S. E. Dosso et al., J. Acoust. Soc. Am. 104, 846–859 (1998)] with excellent results. Also a simple least-squares approach for the solution of the system is implemented; results of the two approaches are compared. Finally, the linearization multipath based technique is successfully applied to real acoustic broadband data for source and receiver localization, and bathymetry and sound speed estimation. [Work supported by ONR.]