If you've spent any time in fitness spaces online, you've seen the headlines: "Free weights aren't better than machines. Science says so."
And honestly, those headlines aren't lying. The meta analyses do show equivalence on certain metrics.
But I think they're answering the wrong question.
They're measuring whether free weights build more muscle or produce more force on a controlled test. The answer is no. But that was never what made free weights essential for athletes. The real question is: which modality builds the nervous system, coordination, and stabilisation capacity that transfers to chaotic, three dimensional sport?
That question has a very different answer. And I can back it up.
What I'll concede up front
I want to start with what the evidence clearly shows, because the strongest position is an honest one.
Hypertrophy is equivalent. When volume and intensity are matched, free weights and machines produce comparable muscle growth. The Haugen et al. 2023 meta analysis (13 studies in total, 5 measuring muscle size) found an SMD of -0.055 (p = 0.751), essentially zero difference, though the authors note the muscle growth evidence is still thin. If your only goal is bigger muscles, a leg press builds quads just fine.
Machines build real strength, on the machine. Machine trained subjects do get stronger. The Haugen meta showed a trend toward machine groups outperforming on machine based tests (SMD: 0.291, p = 0.064). Machines aren't useless. The issue is transfer, not strength.
Machines have legitimate roles in any programme. Volume accumulation when fatigued, training around injuries, isolating weak points, rehab progressions. No serious S&C coach uses free weights exclusively. My argument is about the foundation, not the entire programme.
Now, with that established, here's where the picture changes.
Where free weights actually pull ahead
1. Strength gains are modality specific, and sport is "free weight shaped"
The Haugen et al. 2023 meta analysis found that when both groups were tested on free weight exercises, the free weight groups gained significantly more strength: SMD = -0.210 (p = 0.023). Specificity cuts both ways, though. When both groups were tested on machines, the machine groups tended to gain more (p = 0.064), and for upper body lifts, machine groups gained significantly more on the lifts they trained (p = 0.003). When each group was tested on its own lift, there was no overall difference.
This is the SAID principle (Specific Adaptation to Imposed Demands) operating in real data. You adapt to what you train on. And here's the pivot: sporting actions, sprinting, tackling, jumping, changing direction, all happen in unconstrained 3D space against gravity. That's biomechanically far closer to a barbell squat than a leg press. The specificity data doesn't debunk the free weight case. At minimum, it supports the idea that training adaptations depend heavily on the movement demands you rehearse.
2. In two studies, squat training improved jumping far more than the leg press
For me, this is the most important finding in the debate. Two studies by Wirth et al. (2016), with 39 and 83 people in the training groups, found the same thing: squat training raised jump height by about 12 to 14%, while leg press changes were small and not statistically significant. When Haugen et al. pooled five jump studies, the result leaned towards free weights but did not reach significance (see the chart in section 4).
A third study, Rossi et al. 2018, co authored by Brad Schoenfeld, found the same directional trend (squat CMJ +8.9% vs leg press +3.3%) but the between group differences did NOT reach statistical significance.
Both groups in all three studies got significantly stronger at their own exercise. But the squat groups showed more carryover to jumping, clearly in the two Wirth studies and as a non significant trend in Rossi. None of these studies can tell us exactly why, but the most likely explanation is that the squat movement is closer to a jump.
Worth noting: Neither Wirth study tested both groups on a common strength measure. Each group was only tested on their own exercise. So I can't claim the strength gains were identical. What the data does show is that both modalities built significant strength, but the transfer to jumping was dramatically different.
3. The studies are too short and use the wrong populations
A majority of studies comparing modalities last 6 to 12 weeks and use untrained or recreationally active subjects. Novices improve rapidly on any stimulus. That may hide some differences between modalities. (Haugen 2023 reported approximately 7 untrained and 6 trained study samples, so the short duration criticism is the stronger one here, not the population point alone.)
Sophisticated coordination, stabiliser strength, and sport specific transfer compound over months and years, timeframes these 6 to 12 week trials can't capture. Judging free weights vs machines after 8 weeks with beginners is like comparing a bicycle and a sports car over 10 metres from a standstill. Both cover the distance in about the same time. The difference shows up at 200 km/h on a track.
4. Testing metrics are blind to the dimensions where free weights excel
Standard athletic tests in these studies use bilateral countermovement jumps (sagittal plane, stable surface), isometric dynamometry (single axis, zero degrees of freedom), and linear 30m sprints. Only a small portion of the literature has examined broader athletic outcomes such as reactive agility, loaded multi directional change of direction, dynamic balance, or long term injury patterns in direct free weight versus machine comparisons. (Two small studies are exceptions. Schwarz et al. 2019 tested pro agility, zig zag change of direction and 30 m sprint in recreationally active women: free weight squat, machine squat and drills only groups all improved equally, and the machine squat group gained more jump power. Rossi et al. 2018 tested dynamic balance, and the leg press groups showed slightly larger, non significant, improvements than the squat group.)
When the research says "no difference," it usually means no difference on a fairly narrow set of outcomes that researchers chose to measure. That is useful, but it is not the same thing as testing the full range of qualities that matter in live sport.
5. Less stable, less constrained training can produce higher stabiliser and core activation
A 2024 systematic review and meta analysis (Batista et al.) found that unstable conditions increased activation in several core muscles compared with stable conditions. That does not directly prove free weights beat machines, but it does support the broader point that training environments with greater stability demands create a meaningfully different neuromuscular stimulus.
The internal obliques figure (SMD = 1.04) looks big, but it comes from only two studies of one exercise, and its range of uncertainty is very wide. The steadier core results are small to moderate (0.35 to 0.51). These were mostly body weight exercises on wobbly surfaces, not free weights versus machines, and leg muscle activity didn't change. The one study that compared a free weight squat with a Smith machine squat directly found more calf, hamstring and quad activity with free weights, but no difference in the trunk muscles.
Supporting mechanisms
The Bernstein degrees of freedom problem
Every rep of a barbell squat is a 3D physics problem the brain solves in real time, managing dozens of joints, hundreds of muscles, and thousands of possible trajectories while gravity tries to fold you in half. A Smith machine takes a big part of that problem away by fixing the bar's path. Your muscles still work, but the motor control demands are reduced. And your brain is what runs the show on the field.
This is Nikolai Bernstein's "degrees of freedom" framework from motor control science. A free weight forces the CNS to manage all the variables, balance in three planes, countering gravity, adjusting to perturbations. Machines reduce this complexity by fixing the path of the weight, often to a single line. That makes machines easier. It also makes them a fundamentally different neural stimulus.
Corticomotor plasticity
TMS (transcranial magnetic stimulation) research shows that resistance training enhances corticospinal excitability, increases motor evoked potential amplitude, lowers the stimulation threshold needed to activate the muscle, and reduces short interval cortical inhibition (SICI). Motor skill learning, such as acquiring technique for squats, Olympic lifts, and kettlebell work, produces additional corticospinal adaptations consistent with use dependent synaptic plasticity. Because free weights typically require more motor control than fixed path machines, they may provide a richer corticomotor stimulus. That is a reasonable inference from the motor learning literature, but it has not been cleanly isolated in direct free weight versus machine trials.
Where this leaves us
The meta analyses are not wrong. For building muscle, machines and free weights are interchangeable. For raw force production on a stable, predictable test, they're comparable. I accept that, and any honest coach should.
But athletes don't play sports on a leg press. They play in three dimensions, against gravity, on unpredictable surfaces, with opponents trying to disrupt everything they do. The SAID principle says your body adapts to the specific demands you impose on it. Free weights impose demands that are kinematically, neurologically, and proprioceptively closer to sport than any guided machine.
The squat vs leg press jump data shows it. EMG studies point to a possible mechanism. The specificity findings from the largest meta analysis available suggest that training modality shapes exercise specific adaptation, which strengthens the case that transfer may differ across modalities as well. And the fact that these differences appear even in novices over just 8 weeks, conditions that should minimise the gap, is why I'd like to see longer studies in trained athletes. Until then, this is my expectation, not a proven result.
Use both. But build the foundation with free weights. The research trends, the proposed mechanisms, and common coaching practice all lean in the same direction, even if the full picture is not yet settled.
Sources
Haugen, M.E., et al. (2023). BMC Sports Sci Med Rehabil, 15:103.
Wirth, K., et al. (2016). J Strength Cond Res, 30(5): 1205 to 1212.
Wirth, K., et al. (2016). J Human Kinetics, 53: 201 to 210. PMC5260589.
Rossi, F.E., Schoenfeld, B.J., et al. (2018). J Sports Med Phys Fitness, 58(3): 263 to 270.
Batista, G.A., et al. (2024). Sports, 12(4): 111.
Schwarz, N.A., et al. (2019). Sports, 7(10): 215.
Liang, C. & Liu, H. (2025). Front. Physiol. 16:1569639. doi:10.3389/fphys.2025.1569639
Bernstein, N.A. (1967). The Coordination and Regulation of Movements. Pergamon Press.