As it is the case for any spectroscopic technique, laser-induced breakdown spectroscopy (LIBS) is strongly influenced by the signal fluctuations, and the LIBS spectra need to be normalized to obtain enhanced analytical performance. Nowadays, normalization in LIBS remains an open question and, in the present review, the normalization methods commonly applied to LIBS are presented and discussed, in particular those based on background, total area, internal standard, and Standard Normal Variate. We emphasize that the figures of merit, namely the coefficient of determination, the root-mean square error of prediction and the limit of quantification used to assess the advantages of processing normalized instead of non-normalized LIBS spectra, in a context of quantification, must be calculated in a rigorous way to be able to draw conclusions. We thus propose advices and good practices to achieve a rigorous comparison between quantitative models involving various normalization approaches, the final choice of the best normalization being ultimately driven by the analytical context. In order to take the best advantage from normalization in LIBS and thus increase the analytical performance of this technique, we encourage the analyst to thoroughly compare different normalization methods. Display omitted •LIBS often requires to normalize spectra to reduce fluctuations and increase the prediction ability of quantitative models.•Several normalization methods have been proposed during these last years, but no general rule has been established.•We give the state-of-the-art about normalization methods.•Figures of merit used to assess the performance of quantitative models are provided.•We give advices about the relevance of the figures of merit to robustly evaluate the quantitative model performances.•An example to find the best normalization method for a given analytical context by comparing different model outcomes is proposed.•No normalization method of LIBS spectra can be considered as the best one but different methods should be compared instead.