Seismic interpretation can be viewed as a set of methodologies to enhance the understanding of the structural and stratigraphic model of a given region. During this process, the interpreter analyzes the seismic imaging seeking to identify geological structures such as faults, horizons and channels, among others. Given the low signal to noise ratio, the algorithms that support the interpretation require a pre-processing stage where the noise is reduced. This thesis proposes a new filtering method based on the anisotropic diffusion of the amplitude field. The formulation of the diffusion process proposed here uses seismic attributes to identify horizons and faults that are preserved in the diffusion process. The proposed method implemented in this thesis also presents results applied to real and synthetic data. Based on these results, we present an analysis of the influence of the proposed method in correlation measurements over horizons previously tracked. Finally the thesis presents some conclusions and suggestions for future work.