As part of the intensive efforts in facilitating drug discovery, computational methods have been explored as low-cost and efficient tools for predicting various toxicological properties and adverse drug reactions (ADR) of pharmaceutical agents. More recently, machine learning methods have been applied for developing tools capable of predicting diverse spectrum of compounds of different toxicological properties and ADR profiles. Based on the results of a number of studies, these methods have shown promising potential in predicting a variety of toxicological properties and ADR profiles. This article reviews the strategies, current progresses, underlying difficulties and future prospects in using machine learning methods for predicting compounds of specific toxicological property or ADR profile.