The in vitro phototoxicity assay using the 3T3 cell line is a widely used screening system to identify toxicity risk in humans after exposure to sunlight. The chemical structure of drugs itself is considered to be primarily involved in the process of eliciting phototoxicity. Therefore, various structure-based chemical descriptors have been utilized as a part of in silico prediction model for phototoxicity. Herein, we analyzed the relationship between chemical descriptors and phototoxicity using in-house diverse data sets (2,201 compounds). HOMO-LUMO gaps worked best to discriminate phototoxic and non-phototoxic compounds among the descriptors we assessed. Furthermore, it was revealed that the prediction power was enriched, especially for a subset of compounds not having a fused ring.