Abstract: In order to solve the problem that permeability calculation models have low accuracy for low porosity and low permeability reservoirs, a mathematical model for predicting reservoir permeability has been established by using NMR logging data to characterize spectral area S and cementation exponent on the basis of a study of the representative elementary volume model (REV model). The relationship of each parameter from the REV model and permeability has been analyzed, and the calculation method for all parameters in the model was taken into consideration. The REV model was compared with the classical Timur-Coates and SDR models, the study results indicated that there was a power function relationship between spectral area and permeability, with the correlation coefficient of 0.88. Given a certain of spectral area and porosity, the cementation exponent has a linear relationship with permeability, with a correlation coefficient of 0.82. The value of reservoir permeability is determined by both the spectral area and cementation exponent. There is a good consistency in the change among permeability, spectral area and the cementation index. The study results showed that the accuracy of permeability calculation by REV model was significantly higher than Timur-Coates and SDR models, which is worthy of popularization and application.