A method has been developed to estimate cracking yields for complex hydrocarbon mixtures based on introducing the concept of substitute mixtures of real components. A set of 20 substitute components has been selected according to the true boiling points and densities of atmospheric gas oil distillation cuts. An artificial neural network trained by the full-scale experiments of naphthas cracking provides the prediction of the gas oil cracking yields. The disintegration of individual hydrocarbons into nine structural increments has been proposed for the uniform characterization of naphthas and substitute gas oil. Using the neural network, the yields of gas oil cracking in an industrial reactor can be predicted.