A quadratic Box-Cox methodology is presented for choice of flexible functional form that includes consistent computation of variance estimates. Empirical viability of the procedure is investigated by specifying a dual profit function using highly aggregated U.S. agricultural data. Conditional and unconditional variance estimates for the parameters are compared and contrasted. Like-lihood ratio tests are utilized to discriminate among the generalized Leontief, normalized quadratic, translog, and square-rooted quadratic functional forms. Statistical results indicate that the squarerooted quadratic is the preferred choice of functional form for these data, followed by the normalized quadratic.