The cutoff value of nuclear magnetic resonance (NMR) transversal relaxation time is vital for pore structure characterization, permeability prediction, and irreducible water saturation calculation. Conventional default values often lead to inaccurate results for rocks with complex pore structure. Based on NMR experiments and multifractal theory, we have developed an effective statistical method to predict cutoff values without other petrophysical information. The method is based on multifractal theory to analyze the NMR spectrum with the assumption that the spectrum is an indicator of pore size distribution. Multifractal parameters, such as multifractal dimension, singularity strength, and mass exponent, are calculated to investigate the multifractal behavior of spectrum via NMR experiments and a dyadic scaling-down algorithm. To obtain the optimal cutoff value, the rotation speed and time of centrifugation are enlarged increasingly to optimal centrifugal state. A predicating model for cutoff value based on multiple linear regressions of multifractal parameters was proposed after studying the influential factors. On the basis of the multifractal analysis of NMR spectrum, a reasonable predication model for cutoff value was rendered. Upon testing, the predicted results were highly consistent with the experimental results.