Large-scale proteomic studies have to deal with unwanted variability, especially when samples originate from different centers and/or multiple analytical batches are needed. Such variability is typically added throughout all the steps of a clinical study, from biological sample collection and storage, sample preparation, spectral data acquisition, to peptide/protein quantification. In order to remove such diverse variability, normalization of the protein data is performed. There are several published works comparing normalization methods in the -omics field, but reports focusing on proteomic data generated with mass spectrometry (MS) are much fewer. Additionally, most of these studies have only dealt with small datasets. As a case study, we focused on the normalization of a large quantitative MS-based proteomic dataset obtained with isobaric tandem-mass tagging (TMT) of plasma samples from an overweight and obese pan-European cohort. Different normalization methods were evaluated, namely, standardization, quantile sample, removal of unwanted variation (RUV), ComBat, mean and median centering, and single standard normalization; some of these methods are generic while others have been specifically created to deal with genomic or metabolomic data. We checked how relationships between proteins and clinical variables were impacted after normalizing the data with the different methods. We compared the normalized datasets using an array of diagnostic plots. Some methods were well adapted for this particular large-scale shotgun proteomic dataset of human plasma samples. In particular, quantile sample normalization, RUV, mean and median centering showed very good performance, while quantile protein normalization provided results of inferior quality than those obtained with unnormalized data.