To study the use of electromagnetic scanning in prediction of lean content in beef carcasses and cuts, 100 beef cattle (60 steers and 40 heifers), representing a broad range in external fat thickness (.1 to 2.9 cm) and live weight (414 to 742 kg), were selected. Chilled right sides were divided into streamlined (foreshank, brisket, and ventral plate removed) forequarters (FQ) and full hindquarters (HQ) and scanned. Primal rounds, loins, ribs, and chucks were fabricated from the right side, scanned, and physically separated into lean, fat, and bone. Prediction equations for dissected lean content and percentage of lean included the peak of the electromagnetic scan response curve (obtained from scanning the HQ or FQ), length, temperature and weight of the scanned cut, and fat thickness at the 12th rib. Using the coefficient of determination, root mean square error, and Mallows' Cp statistic, the best model for each dependent variable (weight and percentage of lean) that included up to five independent variables was selected. Prediction equations for the HQ or FQ of steers accounted for 84 to 93% of the variation in lean weight of beef sides and quarters and 71 to 93% of primals. Sixty-one to 75% of the variation in percentage of lean in sides and quarters and 48 to 65% of primals was also explained. Similar results were obtained for heifer carcasses. Predicting percentage of lean in any scanned cut, rather than weight of lean, accounted for less of the variation. Weight and fat thickness contributed significantly when predicting percentage of lean.(ABSTRACT TRUNCATED AT 250 WORDS)