In recent years, mass spectrometry imaging (MSI) has been shown to be a promising technique in oncology. The effective application of MSI, however, is hampered by the complexity of the generated data. Bioinformatic approaches that reduce the complexity of these data are needed for the effective use in a (bio)medical setting. This holds especially for the analysis of tissue microarrays (TMA), which consist of hundreds of small tissue cores. Here we present an approach that combines MSI on tissue microarrays with principal component linear discriminant analysis (PCA-LDA) to predict treatment response. The feasibility of such an approach was evaluated on a set of patient-derived xenograft models of triple-negative breast cancer (TNBC). PCA-LDA was used to classify TNBC tumor tissues based on the proteomic information obtained with matrix-assisted laser desorption ionization (MALDI) MSI from the TMA surface. Classifiers based on two different tissue microarrays from the same tumor models showed overall classification accuracies between 59 and 77%, as determined by cross-validation. Reproducibility tests revealed that the two models were similar. A clear effect of intratumor heterogeneity of the classification scores was observed. These results demonstrate that the analysis of MALDI-MSI data by PCA-LDA is a valuable approach for the classification of treatment response and tumor heterogeneity in breast cancer.