Participants in U.S. markets for live cattle increasingly rely on federal grading standards to price slaughtered animals. This change is due to the growing prominence of "grid" pricing mechanisms that specify explicit premiums and discounts contingent on an animal's graded quality class. Although these changes alter the way cattle are priced, the technology for sorting animals into quality classes has changed very little: human graders visually inspect each slaughtered carcass and call a "quality" and "yield" grade in a matter of seconds as the carcass passes on a moving trolley. There is anecdotal evidence of systematic bias in these called grades across time and regions within U.S. markets. We examine whether such claims are supported in a sample of loads delivered to three different midwestern packing plants during the years 2000-2002. Overall, results indicate that indeed there is a bias, and that grading standards vary significantly across years and packing plants. Results also are consistent with a behavioral model where grader accuracy is inversely related to carcass quality.