This study investigates gains in efficiency from Zellner's Seemingly Unrelated Regressions (SUR) procedure using an optical scanner data base supplied by Information Resources, Inc. A total of 162 sales equations describing different brands, product categories, and cities is estimated with both SUR and ordinary least squares (OLS). The coefficient standard errors obtained from SUR are about 25% smaller than those obtained from OLS. Gains in efficiency from SUR are found to be related to characteristics of the sales models and data bases. Large gains in efficiency are observed for coefficients of equations describing the sales of heavily advertised, high-market-share brands, sold at locations utilizing different marketing policies to attract a similar clientele. These features should be recognized in the data collection and analysis processes.