Some spectroscopies are intrinsically poorly sensitive, such as Nuclear Magnetic Resonance (NMR) and Raman spectroscopy. This drawback can be overcome by using Singular Value Decomposition (SVD) and low-rank approximation to denoise spectra and consequently increase sensitivity. However SVD limits have not been deeply investigated until now. We applied SVD to NMR and Raman spectra and showed that best results were obtained with a square data set in time domain. Automatic thresholding was applied using Malinowski’s indicators. 6×7380 noisy spectra with 41 signal-to-noise ratios were compared to their non- noisy counterparts, highlighting that SVD induces a systematic error for Gaussian peaks but faithfully reproduces shape of Lorentzian peaks, thus allowing quantification. Used carefully, SVD can decrease experimental time by a factor of 2.3 for spectroscopies. This study may help scientists to apply SVD to denoise spectra in a more efficient way, without falling into pitfalls.