In long horizontal wells, production rate is typically higher at the heel than that at the toe. The resulting imbalanced production profile may cause early water or gas breakthrough into the wellbore. Once coning occurs, well production may be severely decreased due to limited flow contribution from the toe. To eliminate this imbalance, inflow control devices (ICDs) are placed in each screen joint to balance the production inflow profile across the entire lateral length and to compensate for permeability variations. Currently, there are three different passive ICD designs in the industry: nozzle-based, helical channel, and tube-type. They use restriction mechanism (nozzle-based), friction mechanism (helical channel) or cooperating the two (tube-type) to achieve a uniform inflow profile. However, the reality is that none of these ICDs alone meets the ideal requirements of an ICD designed for the life of the well: high resistance to both plugging and erosion, high viscosity insensitivity, and high density sensitivity. Therefore, the selection and optimization of ICDs for a specific reservoir requires for further study. In this paper, three computational fluid dynamic based, numerical models of these ICDs with same flow resistance rating (FRR) were developed to characterize the flow performance. The results show that the throttle pressure drop depends on fluid properties, and geometries of each ICD. The weight of each factor that affects pressure drop was determined by maximizing deviations combining with analytic hierarchy process, and subsequently optimal ICD under specific reservoir condition was selected with the help of fuzzy comprehensive evaluation, thereby an ICD selection diagram was built. For a specific reservoir, we will have the selected ICD with best pressure drop composition by optimizing its structural parameter, which has best corrosion resistance and least viscosity sensitivity.