Among the important determinants of the quality of the final models are outlier compounds and the applicability domain of the developed models. Outlier compounds may generally be of two types: structural outliers (sharing chemical characteristics uncommon to most of the remaining compounds) and prediction outliers (showing high residuals, unlike most of the remaining compounds). Identifying outliers is very important, as they may indicate important physicochemical features not yet considered during model development and may be a starting point for exploring a different mechanism of action. Some of the prediction outliers may be, in reality, activity cliffs, which show a significant difference in the activity values with compounds that are pretty similar in structural characteristics. When applying a QSAR model to predict external compounds, it is crucial to verify that they fall within the model’s applicability domain, as no model can be universally applicable to all chemical compounds. The relevance of a QSAR model mainly depends on the chemical domain of the training compounds from which the model has learnt the structure-activity relationships.