The occurrence of harmful algal blooms has been increasing significantly around the world. In order to ensure the safety of drinking water, procedures to screen potential materials as effective algicides are needed, and predictive methods which save both the labor and time compared with traditional experimental approaches, are particularly desirable. In this study, data from previous studies on the algal-growth inhibitory action of two kinds of compounds, namely, the action of fatty acids and thiazolidinediones on the harmful algae Heterosigma akashiwo and Chattonella marina, were modeled using multiple linear regression (MLR) based on quantitative structure-activity relationships (QSAR). The models were shown to have highly predictive ability and stability, and provided insight into the inhibitory mechanisms of congeneric compounds. The main descriptors in the fatty-acid models were the Connolly accessible area and the number of rotatable bonds, illustrating that molecular surface area and shape are important in their algicidal actions. In the thiazolidinedione models, the critical volume, octanol-water partition coefficient (LogP), and Connolly solvent-excluded volume were found to be significant, indicating that hydrophobicity, substituent group size, and mode of action are mechanistically important. Our results showed the algicidal activity of a series of compounds on different algae could be modeled, and each model is efficacious for compounds that fall into the application domain of the QSAR model. This work demonstrates how reliable predictions of the algicidal activity of novel compounds and explanations of their inhibitory mechanisms can be obtained.