Monoclonal antibodies (mAbs) compounded into the hospital pharmacy are widely used nowadays. Their fast identification after compounding and just before administration to the patient is of paramount importance for quality control at the hospital. This remains challenging due to the high similarity of the structure between mAbs. Analysis of the ultraviolet spectral data of four monoclonal antibodies (cetuximab, rituximab, bevacizumab, and trastuzumab) using unsupervised principal component analysis led us to focus exclusively on the second-derivative spectra. Partial least squares-discriminant analysis (PLS-DA) applied to these data allowed us to build models for predicting which monoclonal antibody was present in a given infusion bag. The calibration of the models was obtained from a k-fold validation. A prediction set from another batch was used to demonstrate the ability of the models to predict well. PLS-DA models performed on the spectra of the region of aromatic amino acid residues presented high ability to predict mAb identity. The region corresponding to the tyrosine residue reached the highest score of good classification with 89 %. To improve the score, standard normal variate (SNV) preprocessing was applied to the spectral data. The quality of the optimized PLSDA models was enhanced and the region from the tyrosine/tryptophan residues allowed us excellent classification (100 %) of the four mAbs according to the matrix of confusion. The sensitivity and specificity performance parameters assessed this excellent classification. The usefulness of the combination of UV second-derivative spectroscopy to multivariate analysis with SNV preprocessing demonstrated the unambiguous identification of commercially available monoclonal antibodies.