Quantitative structure–toxicity relationship (QSTR) studies play an important role in toxicity forecasting, which is widely used in the study of modern compounds. Benzoic acid derivatives are an important type of organic chemicals. Most of them may cause serious public health and environmental problems. The effect of quantum-chemistry parameters on the toxicity of benzoic acid derivatives in rats via oral 50% lethal dose (LD50) was studied by QSTR, and a model to predict the toxicity of benzoic acid derivatives was constructed. In order to obtain an accurate model, cross factors were considered and a model for predicting the toxicity of benzoic acid derivatives in rats via oral LD50 was built using a linear regression method (correlation 0.990). The novel model is logtoxi ¼ 0:144 LogP 0:0269SAG þ 0:0000127HoF 0:000377PE, R2 = 0.990, C(p) = 4.000, mean square error (MSE) = 0.785. The model demonstrated good forecasting ability.