This work is concerned with the development of an artificial neural network (ANN), capable of classifying two-phase flow patterns, such as discrete bubbles, stratified, slug-flow, intermittent and annular regimes. Experimental operating data from the literature and the physical properties of the fluids were used to define and calculate dimensionless numbers. These numbers constitute the inputs of the neural model. They successfully describe the flow because they account for the competing forces occurring within the multiphase fluid. The training procedure was performed using a Levenberg-Marquardt algorithm. The methodology used to find the best network architecture is described in detail. All the flow regimes were accurately classified presenting only a small deviation. The final goal is to develop an automatic classification tool for multiphase flow patterns aimed at laboratories and field applications.