Industries / Food, farming and forestry / Food and beverage processing / Meatpacking
Grading carcasses
Grading is the work of measuring each carcass for how much saleable meat it holds and how well it will eat, so that packers can sort it and pay farmers by its value.
7
Milestones achieved, from 1927 to 2025.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1927
US beef grading begins
The US Department of Agriculture drew up tentative grades for dressed beef in 1916 so that market reports could quote prices by grade, and the Army, Navy and Allies used them to buy beef in the First World War. The Secretary of Agriculture made them official on 3 June 1926, and a voluntary service that graded and stamped carcasses began in May 1927.
Credit
USDA, through its Bureau of Agricultural Economics
1976
Shape dropped from beef grades
USDA removed conformation, the shape and muscling of the carcass, from beef quality grades, after research showed it was unrelated to how the meat ate. Every graded carcass had to carry both a quality grade and a yield grade, which estimates the share of saleable retail cuts and was added in 1965 under the name cutability. Announced in 1975, the change was held up by court actions until 23 February 1976.
Lessona trait graders had judged for decades did not predict eating quality, and USDA had first proposed dropping it in 1962.
Credit
USDA
1984
Europe measures pig lean
Council Regulation 3220/84 graded pig carcasses across the European Community by lean-meat percentage, measured with instruments on the carcass, in five grades 5 points apart. The aim it states is fair payment to farmers for the weight and composition of the pigs they deliver; a 1981 regulation had already set the EUROP scale for beef by conformation and fat cover.
Lessoneach country calibrates its own formulas on its own pigs. In 2016 M. Font-i-Furnols and five colleagues applied 12 approved national formulas for one method to the same 951 carcasses and found differences of almost 5 lean-meat percentage points.
Credit
Council of the European Communities
1998
Grades set by consumer taste
Meat Standards Australia (MSA) started commercial grading in 1998, licensing two Queensland processors. Instead of grading by traditional carcass traits, which its developers found "of little value" in predicting what consumers thought, it predicts taste-panel scores, by 2008 for 135 combinations of cut and cooking method on each carcass. MLA says the model rests on almost 1.7 million taste tests by more than 250,000 consumers in 13 countries, and 6.64 million cattle were MSA-graded in 2021–23. In 2017 the UN Economic Commission for Europe endorsed its consumer testing protocols as the global standard.
Lessona 1995 trial with Woolworths assessed 22,591 carcasses with the best meat science of the day, and steaks from stores selling graded beef were no less often unacceptable than the rest, until a further trial changed stimulation and chilling.
Credit
Rod Polkinghorne, J. M. Thompson and the Beef Cooperative Research Centre; Meat & Livestock Australia
2009
Cameras join USDA graders
USDA approved an instrument to measure ribeye area in August 2001, instrument assessment of yield grade in March 2007, and of marbling, the flecks of fat within the lean that set the Prime, Choice and Select grades, in September 2009.
Credit
USDA Agricultural Marketing Service (AMS)
2024
X-rays measure lamb yield
Dual-energy X-ray absorptiometry (DEXA) scans each chilled carcass and reports its lean meat, fat and bone. AUS-MEAT, the body that keeps Australia's meat trade language, must approve each plant's device before farmers are paid on its results, and it accredited DEXA for chilled lamb in December 2024. By August 2026 Meat & Livestock Australia listed lamb DEXA at plants including WAMMCO, JBS Bordertown, JBS Cobram, TFI Tamworth, Gundagai Meat Processors and Frewstal, and Gundagai pays a bonus for lambs that hit yield and eating-quality targets.
Credit
AUS-MEAT; Meat & Livestock Australia; the processors
2025
A phone grades marbling
On 22 August 2025 USDA AMS approved three more vision grading instruments for its beef grading program. One is Global Meat Imaging's system on a Google Pixel 7a phone, approved to predict the official marbling score; the other two also predict ribeye area and yield grade. In Denmark, Danish Crown began classifying marbling with Frontmatec's Q-FOM Beef camera at its Aalborg and Holsted plants, replacing trained classifiers, and pays up to 150 øre (about US$0.22) a kilogram for carcasses that meet its marbling targets.
Credit
USDA AMS; Global Meat Imaging; Danish Crown and Frontmatec
Watch for
Remote grading for small plants
In USDA's Remote Grading Program for Beef, a plant worker photographs each carcass with a phone app and a USDA grader elsewhere assigns the grade. USDA says more than 90% of US fed beef is graded, almost all at large plants, and that the program has registered 88 facilities in 32 states, many of them small, with most grades returned within 24 hours.
Achieved whenan outside party, such as USDA's Inspector General or a university, publishes how remote grades compare with on-site grades on the same carcasses.
Claimed by
USDA AMS, June 2026
Paying beef farmers for yield
James Carter, a processing manager at the Australian Agricultural Company and a 2025 Nuffield Scholar, reports that Australia has never properly measured or paid for beef carcass yield. In his bone-out of 17 carcass sides with Murdoch University, value differences routinely exceeded $200 a carcass, which he scales to a potential $375 million a year across the industry. Teys Australia's Lakes Creek plant was the first Australian beef plant to install DEXA, which takes 48 m² of floor.
Achieved whenAUS-MEAT adds a beef yield trait to its trade language and processors pay on an accredited yield device.
Claimed by
James Carter, Nuffield Australia, August 2026
Microwaves read lamb fat
Meat & Livestock Australia's forum slides say Murdoch University's Microwave Scanner measures intramuscular fat in the lamb loin on the hot carcass at chain speed without cutting it, and that its readings feed automatic sorting in the chiller. The slides' only accuracy figure, "88%+", is not defined.
Achieved whenAUS-MEAT approves the scanner for payment and an outside trial publishes its error against laboratory fat measurements.
Claimed by
Meat & Livestock Australia, for Murdoch University, August 2026
Open
01
Open ribeye photos with grades
A public, openly licensed set of photos of the ribeye face at the 12th rib, each with its official USDA marbling score, ribeye area and grade, from several plants, lighting setups and phone models, with a held-out test split and a baseline model anyone can try to beat.
The briefCredit · unclaimed02
Phone replaces the grading ruler
A phone app for small pig slaughterhouses. It photographs the cut face of a split carcass beside a printed scale, finds the two points of the ZP ("two-point") method, the thinnest fat over the gluteus medius muscle of the rump and the distance from that muscle to the spinal canal, and applies the country's approved formula to give the lean-meat percentage and grade.
The briefCredit · unclaimed
Open 01 · the brief
Open ribeye photos with grades
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, openly licensed set of photos of the ribeye face at the 12th rib, each with its official USDA marbling score, ribeye area and grade, from several plants, lighting setups and phone models, with a held-out test split and a baseline model anyone can try to beat.
Why it is still open
USDA has approved commercial vision instruments, including one on a phone, but their training data and models are the makers' own. Research code shared on GitHub, such as a 2025 project that adapts a marbling model from one plant to others, points to plant images that are not released. GitHub and Hugging Face searches for this file found only small hobby sets without official grades.
Built with what exists
Ordinary phones, a color card in each frame, PyTorch or Ultralytics YOLO for the baseline model, and Hugging Face Datasets for hosting.
First step
Ask a university meat laboratory that has its carcasses graded by USDA to let you photograph 200 ribeyes at grading time, with the grader's scores.
Done looks like
At least 5,000 images from at least three plants and three phone models, released under CC BY with official scores; a baseline trained on two plants and tested on the third, reporting the share of carcasses within half a degree of marbling of the official score and the agreement at the Choice and Select line.
Open 02 · the brief
Phone replaces the grading ruler
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app for small pig slaughterhouses. It photographs the cut face of a split carcass beside a printed scale, finds the two points of the ZP ("two-point") method, the thinnest fat over the gluteus medius muscle of the rump and the distance from that muscle to the spinal canal, and applies the country's approved formula to give the lean-meat percentage and grade.
Why it is still open
In 2013, 11 EU countries had the ZP method approved with a ruler, and the researchers comparing national formulas blamed part of the differences on the tools used to measure ZP and on the operator. Camera versions exist for fixed slaughter lines, such as one CSB-System AG patented in Germany (priority 2004, US grant 2011). A 2024 Chinese study measured backfat in carcass photos with a YOLOv8 model (R² of 0.71 against hand measurement), for breeding, not EU grading. Phone grading has been shown for beef: an app trained on 500 cattle carcasses agreed with expert graders on the EUROP conformation class for 84.2% of them (Negretti and colleagues, 2022). No phone method authorized under EU grading rules turned up, and EU rules let countries exempt plants that slaughter fewer than 500 pigs a week from compulsory grading.
docs.eaap.orgpatents.google.comdoi.orgeur-lex.europa.euebi.ac.uk
Built with what exists
A phone camera, printed ArUco markers for scale, an image segmentation model such as Ultralytics YOLOv8-seg, OpenCV for measurement, and the national ZP formulas published in the European Commission's authorizing decisions.
First step
At a small abattoir that grades by ruler, photograph 50 split carcasses beside a printed scale alongside the grader's ruler readings, and check whether the phone's two distances agree within 2 mm.
Done looks like
A trial set up as EU rules require for authorizing a grading method: carcasses dissected or CT-scanned for reference lean, and a root mean squared error of prediction below 2.5 lean-meat percentage points on a test set of at least 60 carcasses, with two operators photographing each carcass and their results compared.