Based on capillary pressure curve and experimental measurement data of porosity and permeability of 82 carbonate rocks of Zahnanor oilfield in Caspian Seashore Basin，Kazakhstan，comparison analysis of permeability prediction accuracy of carbonate reservoir was made through comparing and analyzing Swanson model，Capillary-Parachor model，Winland model，Pittman model，Nelson model and the δ function model. The results show that the calculation accuracy of the six permeability prediction models is poor for permeability prediction of complicated pore structure carbonate reservoir. But the effect of Swanson model is better than those of the other five models. Further analysis on the applicability of permeability prediction of carbonate reservoir based on Swanson model suggestes that the Swanson model cannot reflect the influences of microfracture and complex pore throat distribution on permeability. For carbonate rock sample，the pore throat，the distribution of which is characterized by multimodal feature，contributes to reservoir permeability，and there is no dominant pore throat size in the core sample. On the contrary，the Swanson model believes that the pore throat radius corresponding to the inflexion point of capillary pressure curve is the dominant pore throat size of core sample. And the Swanson model ignores the influence of small pore throat on permeability of rock sample which has a bimodal pore throat distribution. Therefore，the error of permeability calculated by Swanson model is obvious. Based on Swanson model，we put forward an improved permeability prediction model considering the influence of pore structure parameters such as porosity，skewness and sorting coefficient. It is found that the permeability prediction accuracy of the improved permeability prediction model for complex pore structure carbonate reservoir is obviously enhanced through comparing it with the Swanson model.