This study attempts to develop the original Armington procedure into a multi-CES non-homothetic model. The results using rice trade data indicate that the new approach is superior to the original. Elasticities of substitution for rice imported from individual suppliers appear to vary, and demand for imported rice is not homothetic.