Obtaining accurate T2 cutoff value is very important for reservoir evaluation using NMR data. Traditional T2 cutoff value acquisition method is obviously affected by regional actual situation and reservoir heterogeneity, and often has large error in practice. Based on core NMR experiments, this paper analyzes the relationship between physical parameters, Geometric mean of T2 spectrum in NMR and T2 cutoff values. Considering the influence of reservoir heterogeneity, a graph theory multi-resolution clustering method is proposed to classify logging facies by using neighboring index (NI) and kernel representative index (KRI). The corresponding relationship between logging facies and lithofacies is established based on casting thin section and other data, and a prediction model of variable T2 cutoff value based on logging facies constraint is established. The practical application of X oilfield in Bohai Basin shows that the error between the T2 cutoff value calculated by the proposed model and the T2 cutoff value measured by core NMR experiment is small, which has obvious advantages over the traditional method, and further improves the accuracy and applicability of using NMR data to calculate reservoir parameters.