The work presents a method for characterizing underwater acoustic signals using a Markov chain, with hidden variables, based on their wavelet transform. Initially, we assign to the signal a Hidden Markov Model (HMM) for which the conditional posterior probability density function seems to be the most representative using an Expectation-Maximization algorithm. Special techniques are applied to avoid over-fitting which in principle is not desirable for the sought applications. The features used for the assignment consist of two dimensional time series obtained by preprocessing of signal’s wavelet packet coefficients. Subsequently, we use an approximation of the Kullback Leibler (KL) divergence as a similarity measure among the HMMs. The approximation is obtained by employing Monte-Carlo (MC) techniques simulating the significant sampling from the HMMs posterior distributions. This technique is used in cases where the similarity of two or more signals is to be exploited. These cases include a variety of problems associated with the monitoring of the marine environment using acoustic or seismic signals. The applications to be presented here are referred to problems of geoacoustic inversions (seabed mapping) using simulated acoustic data and seismic monitoring using real data from a terrestrial seismograph to illustrate the various possible applications of the suggested method.