Upper Back Exercises: Rhomboids, Mid Traps and Thickness

What counts as the upper back

Upper back exercises train the muscles that move and steady your shoulder blades: the rhomboids, the middle trapezius and the lower trapezius. The rhomboids and middle traps pull the shoulder blades back toward the spine, and the lower traps draw them down and help rotate them as your arms rise. Lifters call this region the mid back, so rhomboid exercises and mid back exercises are largely one list.

The upper traps belong to the same muscle but mainly shrug your shoulders toward your ears; for those, use trap exercises. The rear delts and lats help in every row and have their own guides: rear delt exercises and lat exercises. For the whole back, read what the research says about the best back exercises or browse the back exercise hub.

The upper back exercises worth your sets

No growth trial has measured the rhomboids or the middle and lower traps, so this ranking puts mechanism and loadability first and EMG second. Rows lead because they move the shoulder blades through a full range under heavy load; raises and face pulls follow as lighter work.

Chest support earns first place on mechanism: the pad holds your torso, so the set ends when your upper back tires, not your lower back. In a biomechanics study of 7 men, the standing bent-over row loaded the lumbar spine most and the inverted row least, and the inverted row also gave the highest upper-back activation of the three. An EMG study of 15 trained men found a two-arm free-weight row raised lower-back muscle EMG by 34% over a two-arm machine row.

The seated cable row is a close second. In an EMG study of 12 trained men, the seated row produced the highest signal at a middle trapezius and rhomboid electrode site, ahead of wide-grip and reverse-grip pulldowns. Use T-bar and barbell rows when your lower back recovers well, and swap to rank 1 when it does not.

Face pulls, band pull-aparts and the reverse pec deck have no data in this guide. They earn a place on mechanism, since the arms pull back and apart while the shoulder blades retract. For dumbbell-only training, see dumbbell back exercises.

What the scapular muscle studies show

These are EMG studies, which record electrical activity during a few reps, and a 2022 review argues that the size of a surface EMG signal has not been validated as a predictor of muscle growth. Surface electrodes also cannot isolate the rhomboids: an EMG methods study in 21 healthy people classes the rhomboid major as a deeper-lying muscle recorded with fine-wire electrodes, so surface sites labeled middle trapezius and rhomboid pick up a combined signal.

Prone raises lead the data for the middle and lower traps. An EMG study of 30 healthy people found that lying face down and raising straight arms out to the side with the thumbs up, or overhead in a Y, produced the highest middle-trap EMG, and the overhead Y produced the highest lower-trap EMG. A lab study of 45 healthy people found the thumbs-up T and a prone arm extension activated the middle and lower traps with relatively little upper-trap activity.

Rows hold up in the same kind of data. In an EMG study of 17 women, the one-arm row was among the exercises that activated the middle trap more than the upper trap. An indwelling-electrode study of 9 people, one of the few to record the rhomboids directly, listed rowing among four core scapular strengthening exercises.

These studies used light rehab loads in healthy people and measured no growth. Use them to decide where raises and rows fit, not to promise thickness. Upper-trap activity is not something to avoid; these exercises simply shift more of the work to the middle and lower parts.

Rows for thickness: elbow height and the squeeze

Where your elbows travel shapes which back muscles a row favors. Tuck the elbows and the upper arm swings down toward the hip, so the lats lead. Let them travel out at about 45–60 degrees from your torso and the upper arm moves back and slightly out, so the shoulder blades must retract to finish, shifting work to the middle traps, rhomboids and rear delts.

The classic cue is to squeeze the shoulder blades together at the end, and its support is thin. In the 12-man EMG study above, actively retracting the shoulder blades did not change middle trapezius and rhomboid EMG. A crossover study of 14 trained men found fixed and free shoulder blades gave broadly similar back EMG, with modestly higher rear-delt signal on the pull and higher mid-trap and lat signal on the lowering when the blades stayed fixed.

So let the shoulder blades move: reach forward under control at the start, pull them back as the elbows pass your torso, and pause about a second if it helps you own the position. Use the full range too. In an EMG study of 16 trained men doing prone barbell rows, top-half and bottom-half reps shifted lat and mid-trap EMG only modestly and inconsistently, so partial rows offer no clear EMG advantage.

Back width vs back thickness

Width and thickness are coaching words, not outcomes any study has measured. Width usually means how far the lats flare at your sides, seen from the front or back. Thickness means the depth of muscle around the shoulder blades and along the spine, seen from the side.

The split still helps you choose exercises. Vertical pulls and elbow-to-hip rows bias the lats, while elbows-out rows and prone raises bias the upper back. Every row trains both to a degree, so pairing one lat exercise with one of the back thickness exercises above covers the whole back.

Sets, reps and weekly volume

A 2017 meta-analysis of 15 studies found that more weekly sets per muscle went with more growth in a graded pattern. A 2026 meta-regression of 67 studies points the same way, with diminishing returns as sets climb, and finds little evidence that more training days add growth when weekly sets match.

Start with about 6 direct upper back sets a week if you are new and about 10 if you already row regularly. Count elbows-out rows, prone raises, face pulls and pull-aparts as full sets, and pulldowns and pull-ups as half sets. That half-set rule is the indirect-set method the 2026 meta-regression found fit its data best, a bookkeeping convention rather than a measured value for the rhomboids. The sets per muscle guide explains the counting.

Stop rows 1 to 2 reps short of failure, while your torso still holds its position. Going all the way to failure is optional: a 2022 meta-analysis of 15 trials found it was not needed for growth or strength, and it did not hurt either. Light raises and face pulls can go closer to failure, since the loads are easy to control; the RPE calculator helps you rate how close you are.

• Chest-supported dumbbell row: 3–4 sets of 8–12.
• Seated cable row: 3 sets of 10–12.
• T-bar or barbell row: 3 sets of 6–10.
• One-arm dumbbell row: 3 sets of 8–12 per side.
• Prone Y and T raises: 2–3 sets of 12–20.
• Face pull or band pull-apart: 2–3 sets of 15–25.
• Inverted row: 3 sets of 8–15.

Technique cues for the top picks

Chest-supported dumbbell row

• Set an incline bench to about 30–45 degrees and lie chest-down.
• Let the dumbbells hang and the shoulder blades spread apart.
• Pull the elbows back and out at 45–60 degrees until the upper arms pass your torso.
• If your chest lifts off the pad, the weight is too heavy.

Seated cable row

• Use a wide bar or wide neutral handle, knees soft, torso upright.
• Reach forward at the start without rounding your lower back.
• Pull toward your lower ribs with the elbows slightly out, then return slowly.

Barbell or T-bar row

• Hinge until your torso is about 30–45 degrees above horizontal, back flat.
• Brace before each rep and end the set when your torso starts to rise.
• Pull to your upper abdomen with the elbows angled out.

Prone Y and T raises

• Lie face down on an incline bench, with light plates or none.
• For the T, raise straight arms out to the side with the thumbs up.
• For the Y, raise them overhead in a Y, thumbs up, and pause for a second.

Common mistakes

• Tucking the elbows on every row when thickness is the goal.
• Heaving the torso on barbell rows so the hips move the weight.
• Holding a long, forced squeeze that costs reps without adding measured activity.
• Skipping the reach at the start, so the shoulder blades never travel.
• Going heavy on raises and swinging the weight up.
• Training through sharp pain instead of seeing a qualified clinician.

Regressions and progressions

Add reps first, then the smallest load jump your equipment allows; the progressive overload calculator turns your last session into a next target. At home, the inverted row, band pull-apart and dumbbell rows cover the main patterns.

A sample upper back workout

This upper back workout holds 10 direct sets. New lifters can do the first two exercises only.

Do it once a week, or split it over two upper-body days. On a pull day, put a vertical pull first, then these rows, then the raises; the back and bicep workout pairs a pulldown and a supported row with curls and face pulls. Trainer tracks weekly sets per muscle, so you can watch upper back and lat volume side by side.

Upper back pain: where this guide stops

This guide is about building muscle, not treating pain. If your upper back hurts, pain spreads into an arm, you notice numbness or tingling, or pain started after a fall or injury, see a qualified clinician before you train the area. Do not train through sharp pain, and do not read any exercise here as a treatment.

What the research cannot tell you yet

• The rhomboids lie deep and are recorded with fine-wire electrodes; surface sites labeled middle trapezius and rhomboid record a combined signal, as the methods study describes.
• No growth trial has measured the rhomboids or the middle and lower traps.
• Back width and thickness are coaching language, not outcomes any study has measured.
• Shoulder-blade cues in rows have only acute EMG support. The 2004 EMG study found a deliberate squeeze did not raise middle-back EMG further, and the 2025 crossover found similar overall EMG either way.
• The prone raise data come from light rehab loads in healthy people, so heavier-load behavior is unknown.

Until better data arrives, choose rows and raises by mechanism and load them progressively. Your log, not a cue, will show what keeps working.