This study introduces an accurate, generalized method based on a three-layered perceptron artificial neural network (ANN) trained to estimate the continuous operation of a vacuum gas oil hydrotreater. The ANN models predict the fractional composition of paraffins, naphthenes, and aromatics distribution (mono, di, tri and tetra aromatics) contained in a vacuum gas oil (VGO) mixture. Moreover, ANN models were built to simulate the behavior of several process variables such as product flowrate, average reactor temperature and product quality (sulfur content, API gravity, TBP50, metals content, and refractive index). VGOs proceeding from commercial hydrotreaters were employed to illustrate the use of ANN models to estimate the VGO composition in a very simple manner with no need for knowledge of the system. The artificial neural net models use API gravity, TBP50 and the refractive index of the desired vacuum gas oil to be studied. These empirical models were tested on experimental data not used during training showing an average error of 0.5 % in the prediction of VGO paraffins, naphthenes and total aromatics content, which is an outstanding performance compared with traditional correlations proposed in the literature. The hydrotreater reactor ANN models were compared against process data collected from the CRP-Amuay refinery data acquisition system. The results using the ANN model to estimate the reactor EIT (Equivalent isothermal temperature) indicate an average error of 4 °F. Additionally, the product sulfur content prediction reports an average error equal to 6 wt%.