Abstract: The artificial neural network methods are used in building of an interpretative model for reservoir permeability in order to improve the accuracy of permeability interpretation. Meanwhile, the BP network is improved and an application program for self - checking genetic neural networks is written to more rapidly acquire accurate and stable interpretative models. A permeability interpretative model is built by using the modified artificial neural network to perform level - by - level interpretation of reservoir permeability profile. Some new concepts that can describe more accurately the heterogeneity of reservoirs, including rhythmic dimensionality, rhythmic angle and rhythmic height, are presented in combination with fractal geometry.