A multicrop model of Canadian demand for US crops is estimated with Zellner's seemingly unrelated regression (SUR), which corrects for the distortion problem in contemporaneous correlation, and with ordinary least squares (OLS), which ignores the problem Comparing inference parameters, trade elasticity estimates, and out-of-sample foremast performance of the Canadian import demand model's SUR and OLS versions demonstrates the importance of addressing contemporaneous correlation, even though both estimators are unbiased This article addresses three shortcomings of the agricultural trade literature frequent failure to account for easily corrected econometric problems, excessively wide ranges of trade parameter estimates, and frequent failure by researchers to validate models beyond the sample