This work investigates the use of Artificial Neural Networks (ANN) on the inferential modelling of properties of petroleum products. Inferential modelling aims to provide a good estimation of chemical properties of petroleum products (i.e: final boiling point, vapour pressure). These properties can be determined by on-line process analysers or laboratory analysis. However, these systems provide neither systematically good results nor the necessary frequency to allow control of the process in real time. However if a good estimation of a property of interest is available, it can be used to achieve the control or the optimisation of production process. This work is subdivided in four main sections: (1) a study about the inference of properties of products in a distillation column; (2) a study about the main methods used on inferential modellind and data analysis, with emphasis on ANN; (3) a systematic about development and testing of inference models; (4) and a case study. In the study about principal methods used on inferential modelling involved a bibliographic reserch about the linear regression techniques Multiple Linear Regression (MLR), Principal Component Regression (PCR) and Partial Least Squares (PLS), a semi empirical model and ANNs. Although the main objective of this work was to evaluate the ANNs perfonmance, the study of other methods was important to compare the results. In addition to the many modelling techniques, some other techniques of data analysis were studied, like Principal Component Analysis (PCA). In the systematic about the development and testing of models, the various problems encontered and the approach used to develop and test the model were presented. An environment of development and testing was also implemented in order to provide a platform to produce and test inferential models. The environment can work with all models studied, and some important settings of the models can also be modified. Many capabilities fo MATLAB software were used on the environment. For the development of the case studies, real data gathered from refineries of Petrobras group were used. Three distinct cases were analysed: the first and second cases are models of kerosene (jet fuel) and diesel ASTM distillation; the third is a model of the Liquefied Petroleum Gas (LPG) 95% boil-off point. In all cases, the influence of each input over the modelled variable was analysed, using mainly the PCA technique. Many ANN arquitetures were tested, comparing them with other studied techniques. The developed ANN models achieved good performance, with better results than the statistical methods. It was also verified the influence of pre- processing and statistical analysis on the success of the modeling. Chemical and Petrochemical process industries have used ANNs in many areas. In the field of inferential modelling of properties, the ANNs allow the accomplished of inferential models in a broad and accurate way. It may be used either for control in real time in single control loops or as part of a multivariable controller.