Among the statistical technique utilized by the Centro Docimologico to as-sess the validity of the selection tools of the students entering university, there are also the connessionist models, the neural networks. One of the main aims of the neural networks is to estimate the parameters of a nonlin-ear regression model. The neural model, after a first step of training by means of a set of patterns, is able to forecast the new values (generaliza-tion). In this work we relationed the students’ academic performance, reg-istered in Motor Sciences study course, with the results they achieved dur-ing the selection test using a multilayer feedforward neural network. The forecasting skill of the model has checked using some patterns we previ-ously excluded from the training sample.