Irreducible water saturation, affecting the productivity of a reservoir, is one of the key parameters for a tight sandstone reservoir. Nuclear magnetic resonance (NMR) logging, detecting the relaxation information on pore fluid with hydrogen nuclei in formation, can be used to distinguish irreducible and movable fluids as a result of their different relaxation information. The irreducible water saturation can be determined by the T-2 cutoff method in the conventional reservoir. However, a tight sandstone reservoir possesses a complex pore structure, and relaxation signals of different pore fluids may overlap in the T-2 distribution. The T-2 cutoff method to determine irreducible water saturation may be a challenge in tight sandstone reservoirs. In this study, a novel T-2 distribution-based method, with the film model assumption, was proposed to determine the irreducible water saturation in tight sandstone reservoirs. The T-2 distribution was transformed by a continuous wavelet transform to obtain the two-dimensional matrix of wavelet coefficients, and the position and shape parameters of the spectral peak at different scales in wavelet space were determined. These parameters were used to construct the Gaussian distribution functions (GDFs) at different scales. The T-2 distribution was decomposed on the basis of the constructed GDFs, and then the optimal decomposition was selected. The irreducible water saturation is determined by the optimal T-2 distribution decomposed. The effectiveness and practicability of the proposed method were validated by numerical simulation, core experimental, and NMR logging data processing.