Nuclear magnetic resonance (NMR) has been widely used to evaluate the pore size distribution (PSD) properties of coals. However, the NMR signal itself only provides a relative distribution of pore size. To calculate an absolute pore size distribution from the NMR data, the T-2 cutoff value needs to be known. Meanwhile, the T-2 cutoff value is an indicator to divide the irreducible fluid and movable fluid in porous rock and a key factor for permeability prediction. Conventionally, the T-2 cutoff value is obtained by centrifugal experiments, the process of which is complicated and time consuming, and some T-2 cutoff value prediction models are not suitable for coals with complex pore structures. Hence, a new method is needed for T-2 cutoff value prediction. In this study, we performed scanning electron microscopy (SEM), low-temperature nitrogen adsorption/desorption (LTNA) and NMR experiments on 12 coal samples. The results of SEM and LTNA reveal the complex pore structures of the coals. According to the results from centrifugal experiments, the T-2 cutoff value is in the range from 0.5-2.8 ms for subbituminous coals, whereas it is 15-32 ms for anthracite coals. We present a fractal theory based method for T-2 cutoff value prediction. Using the estimated T-2 cutoff values, we calculated the movable porosities, PSD and permeability for the selected coals. The results show that the proposed permeability model provides significantly better permeability estimation than classic (Coates and SDR) models. These methods are applicable not only for coals, but also for other unconventional gas reservoir rocks such as gas shales.