This article intends to design a non linear model for analysis and prediction of the time-series concerning the external price of Colombian coffee using rtificial Neural Networks. The main purpose is to compare the realm and limitations of proposed model and the linear prediction classic model ARIMA.