This work proposes a method for pore pressure prediction in reservoir rocks through compressional- and shear-velocity data (seismic attributes). In the method, the attributes are considered observations of a physic system, which behavior depends on a several not-observable parameters, where the pore pressure is only one of these parameters. To estimate the pore pressure, a Bayesian inversion approach is adopted. Through the use of a likelihood function, settled through a calibrated rock physics model, and through the Bayes theorem, the a priori information about the not-observable parameters (fluid and rock parameters and stress state) is combined with the seismic attributes, inferring probabilistically the pore pressure. Due the non-linearity of the problem, and due the uncertainties analysis demanding, an algorithm based on Monte Carlo simulations (a special case of the Metropolis- Hastings algorithm) is used to solve the inverse problem. The application of the proposed method is simulated through some synthetic examples. It is shown that a successfully pore pressure prediction in reservoir rocks depends on a set of factors, as how sensitive are the rock velocities to pore pressure changes, the a priori information about rock and fluid parameters and the uncertainties associates to the seismic attributes. Since the current methods for pore pressure prediction use exclusively the attribute compressional velocity V(p), the contribution of the attribute shear velocity V(s) on prediction is evaluated. In a poorly consolidated rock scenario (or in sands), the V(s) data, even with great uncertainties associated, can significantly contribute to a better pore pressure prediction. In a consolidated rock scenario, the uncertainties associated to pore pressure estimates are higher, and the s V data does not contribute to pore pressure prediction as it contributes in a poorly consolidated rock scenario.