The structure-activity landscape of chemical compounds may not always be smooth, though this is a fundamental assumption of QSAR modeling. There may be several cliffs in the landscape, where some chemical compounds may show different activity (of the order 100-fold or more) than other close congeners in the series. In such cases, the predictions for such compounds derived from QSAR would naturally be disappointing, as the similarity principle is not followed here. Thus, identifying activity cliffs is crucial for developing statistically acceptable QSAR models. The definition of similarity for identifying activity cliffs may be based on chemical fingerprints or descriptors (classical activity cliffs), substructures (chirality cliffs, matched molecular pair cliffs), three-dimensional structure-based (3D-cliffs), or the target-set dependent potency difference. Some prediction outliers, even within the applicability domain of QSAR models, may arise due to the activity cliff (AC) behavior. In addition to compound pairs, activity cliffs can also be visualized in coordinated networks that form AC clusters. Despite using high-quality data, the data set’s modelability may be significantly compromised in the presence of ACs, among other factors. Different methods for identifying activity cliffs have been proposed, such as the structure-activity landscape index (SALI), structure-activity relationship index (SARI), and machine-learning-based methods.