The objective of this study was to test and validate electromagnetic scanning of whole pork carcasses in an on-line, integrated, industrial configuration. The electromagnetic (EM) scanner was installed in two pork processing facilities (Plant A and Plant B). Plant A was a small pork fabrication plant that further processed chilled pork carcasses. Carcasses were delivered to Plant A by refrigerated trucks. The amount of EM energy absorbed by the carcasses was recorded as they were conveyed through the EM field. A plot of the absorption units over time (EM scan curve) was used to obtain predictive variables for estimating carcass and primal cut composition. Forty-eight whole, chilled carcasses (Group A) were electromagnetically scanned and conveyed onto the fabrication line. The average percentage carcass lean for Group A was 49.1% (range = 36.5 to 59.5%). Right carcass sides were removed from the processing line, fabricated into primal cuts, and dissected into fat, lean, and bone. Prediction equations were developed from EM scans for weight of total dissected carcass lean (R2 = .830; root mean square error = 1.80 kg), percentage of carcass lean (R2 = .820; root mean square error = 2.29%), and weight of dissected ham, longissimus muscle, and shoulder lean. In Plant B, the electromagnetic scanner was installed at the end of a pork slaughter line to ensure carcass scanning at a consistent carcass temperature. Fifty whole, pre-rigor eviscerated carcasses (Group B) were electromagnetically scanned before entering the chill cooler where fat and loin tissue depths were obtained by an optical grading probe. The average percentage carcass lean for Group B was 46.7% (range = 30.1 to 57.3%). Prediction equations were developed from EM scans for weight of total dissected carcass lean (R2 = .904; root mean square error = 1.59 kg), percentage of carcass lean (R2 = .863; root mean square error = 2.05%), and weight of dissected ham, loin, and shoulder lean. Statistical equations developed for the prediction of dissected primal cut lean were superior from EM scans of Group B (prerigor) carcasses. Electromagnetic scanning proved more statistically efficient than optical probes for predicting weight of dissected carcass lean and percentage of carcass lean. Statistical comparison of EM scan equations from Groups A and B are not completely valid because two different populations of carcasses were tested at different times of the year. The results of this study show that EM scanning has the potential to accurately predict pork carcass composition in a fully automated, on-line industrial configuration.