A muiltiregional Armington model of U S cotton exports is estimated inappropriately with ordinary least squares (OLS) and appropnately wlth seemingly unrelated regression (SUR) Trade elasticity estimates and out-of-sample forecast performance demonstrate the importance of using the correct econometric technique The choice of estimator clearly Influences the model's forecast accuracy out of sample, levels of trade parameter estimates, and degrees of coefficient estimate efficiency Four shortcoimngs of the agricultural trade literature are addressed (1) frequent neglect of trade theory, (2) excessively wide ranges of trade parameter estimates, (3) frequent misuse of OLS, and (4) failure to validate models out of sample