Low-field nuclear magnetic resonance (NMR) technology has been widely used in laboratory core analysis and formation evaluation since the recording signal is only sensitive to the number of hydrogens in fluid, and the rock matrix has no contribution to the NMR response. However, it is a challenge to obtain accurate porosity from NMR logging in tight sandstone formations due to the effect of echo spacing (TE) on short transverse relaxation time (T2) fluid components in T2 distribution. The aim of this paper is to establish an empirical method to correct NMR porosity of tight sandstone by using a multi-TE laboratory experiment, a dual T2 cutoff model, and graph-regulated nonnegative matrix factorization. The effect of TE on the transverse relaxation spectrum is investigated and corrected for different relaxation ranges, which are determined by the dual T2 cutoff model. The graph-regulated nonnegative matrix factorization is adopted to separate the relaxation components from the T2 spectrum, aiming to eliminate the effect of hydrocarbon. This study demonstrated that the corrected NMR porosity agrees well with the laboratory data. This method can be easily extended in other tight sandstone formations to realize NMR porosity correction.