Molecularly imprinted polymers (MIPs) have been used as useful sorbents in solid-phase extraction for a wide range of molecules and sample matrices. Their unique selectivity can be fine-tuned in the imprinting process and is crucial for the extraction of macromolecules from complex matrices such as serum. A relevant example of this is the application of MIPs to peptides in diagnostic assays. In this article the selectivity of MIPs, previously implemented in a quantitative mass-spectrometric assay for the biomarker pro-gastrin-releasing peptide, is investigated. Partial least squares regression was used to generate models for the evaluation and prediction of the retention mechanism of MIPs. A hypothesis on interactions of MIPs with the target peptide was verified by ad hoc experiments considering the relevant peptide physicochemical properties highlighted from the multivariate analysis. Novel insights into and knowledge of the driving forces responsible for the MIP selectivity have been obtained and can be directly used for further optimization of MIP imprinting strategies.