Using the monthly data for beef exports from the United States to Mexico between January 2000 to December 2012 and based on the Box- Jenkins methodology, an autorregressive moving average (ARMA) model was constructed to forecast the behavior of imported beef. Based on the correlogram and the Dickey-Fuller test on the time series, no evidence of non stationary behavior was found, the process was continued using ordinary least squares to estimate the parameters of a group of models with different autorregressive  AR(p) and moving average MA(q) combinations until two were selected: a AR (1) and a ARMA (1, 2). To establish which model better represented the data generating process, the regression coefficients together with the Akaike and Schwarz criteria were used, to determine the presence of autocorrelation both the correlograms were inspected and the Ljung-Box test applied, the result of this evaluation lead to the selection of the AR (1) model, however after testing the forecasting ability of both models it was found that the ARMA (1, 2) model had a better fitting. Although the predicted values overestimated the real values it was concluded that this type of time series model may be considered as a useful tool to short term forecast Mexican imports of US beef.