In silico screening is routinely used in the drug discovery process to predict whether each molecule in a database has a function of interest, such as inhibitory activity for a target protein. However, drugs generally have multiple functions including adverse effects. In order to obtain small molecules with desirable physiological effects, it is useful to simultaneously predict as many functions as possible. We employed Support Vector Machine to build classification models for 125 molecular functions, derived from the MDDR database, which showed higher kappa statistics (0.775 on average) than those of predictions by Tanimoto similarity (0.708). By analyzing the patterns of the predicted values (functional profiles) of 871 marketed drugs, we demonstrated its applications to indication discovery, clustering of drugs, and detection of molecular actions related to adverse effects. The results showed that functional profiling can be a useful tool for identifying the multifunctionality or adverse effects of small molecules.