This work presents L1/L2 two-parameter regularization as an efficient technique for the identification of light oil in the two-dimensional (2D) nuclear magnetic resonance (NMR) spectra of tight sandstone reservoirs. A 2D NMR T2-T1 distribution model containing light oil, natural gas, and formation water is constructed. 2D NMR echo trains are obtained by means of the multiwaiting time Carr, Purcell, Meiboom, and Gill pulse sequence. A detailed analysis of the ill-posed characteristics to obtain 2D NMR spectrum is given using a Picard curve. The identification abilities of L1/L2 two-parameter regularization and three other techniques are compared in detail. The paper demonstrates that even if the signal-to-noise ratio (SNR) is around 100, it is still very difficult to obtain the 2D NMR spectrum. Light oil cannot be distinguished using Tikhonov regularization and truncated singular value decomposition. Both L1/L2 two-parameter regularization and L1 norm regularization can identify light oil, while the identification ability of L1/L2 two-parameter regularization is higher than that of L1 norm regularization, especially when the SNR is very low. L1/L2 two-parameter regularization can be used as an effective technique to identify light oil from formation water in the 2D NMR spectra of tight sandstone reservoirs.