Drug hypersensitivity (DH) reactions are clinically unusual because the underlying immune stimulations are not antigen-driven, but due to non-covalent drug-protein binding. The drugs may bind to immune receptors like HLA or TCR which elicits a strong T cell reaction (p-i concept), the binding may enhance the affinity of antibodies (enhanced affinity model), or drug binding may occur on soluble proteins which imitate a true antigen (fake antigen model). These novel models of DH could have a major impact on how to perform risk assessments in drug development. Herein, we discuss the difficulties of detecting such non-covalent, labile and reversible, but immunologically relevant drug-protein interactions early on in drug development. The enormous diversity of the immune system, varying interactions, and heterogeneous functional consequences make it to a challenging task. We propose that a realistic approach to detect clinically relevant non-covalent drug interactions for a new drug could be based on a combination of in vitro cell culture assays (using a panel of HLA typed donor cells) and functional analyses, supplemented by structural analysis (computational data) of the reactive cells/molecules. When drug-reactive cells/molecules with functional impact are detected in these risk assessments, a close clinical monitoring of the drug may reveal the true incidence of DH, as suppressing but also enhancing factors occurring in vivo can influence the clinical manifestation of a DH.