Forecasting the demand for cash in Colombia has become a true challenge in the recent past.The last decade witnessed strong changes in the variables that determine the demand for money: Inflation and, hence, interest rates, fall substantially, technological progress was strong in the Colombian Payment System and distorting Tobin-like taxes to financial transactions were imposed.These changes are of special relevance when the demand for money is a non-linear function of its determinants.In this paper we exploit the flexibility of artificial neural networks (ANN) to explore the existence of nonlinearity in the demand for cash.The results show that the ANN models outperform those of linear nature in terms of forecast errors.Furthermore, significant evidence is found of non-linearity in the dynamics of the demand for cash.