The electrotopological state (E‐state) index was employed to characterize the structures of 51 substituted benzenes. Eleven E‐state indices of the compounds were calculated by the computer program developed in our laboratory. The method for variable selection and modeling based on prediction (VSMP) was used to select an optimal combination of the variables from 11 E‐state descriptors. Then the optimal descriptors were employed to model the relationship between the relative biodegradability of the substituted benzenes and their molecular structures. A novel 5‐descriptor linear model was developed and the model has a high quality with the correlation coefficient and the root mean square error in estimation step being 0.9378 and 0.35, respectively, and these in leave‐one‐out cross‐validation procedure being 0.9210 and 0.39, respectively.