Do you have to lift heavy to grow?

One question that is really two

The standard advice — heavy loads for strength, six to twelve repetitions for size, light loads for endurance — treats muscular adaptation as a single dial. It is not. Growing a muscle and raising the maximum weight you can move once are distinct outcomes that respond to load differently. Asking "do I have to lift heavy?" without saying heavy for what gets an answer that is half right.

One condition has to be held fixed before the comparison means anything, and most gym arguments quietly drop it: the sets have to be hard. A set of fifteen with a weight you could have taken for thirty is a warm-up, not a light-load set.

When both loads are taken to failure

Schoenfeld and colleagues pooled trials comparing low-load training (≤60% of 1RM) with high-load training (>60% of 1RM) under one strict rule: every set in every protocol had to be taken to momentary muscular failure.¹ Twenty-one studies met the criteria.

The results split by outcome. Both loading conditions improved every outcome; the question was whether one improved them more. The authors attribute the divergence between the two strength measures to specificity: a 1RM test rewards practice at the tested load, whereas isometric force is read on a neutral instrument, which offsets that advantage and shrinks the gap.

In percentage terms the same asymmetry appears. Heavy training raised 1RM by 35.3 ± 4.3% against 28.0 ± 4.8% for light; muscle size rose 8.3 ± 1.5% against 7.0 ± 1.2%. The authors' own summary is that maximal strength benefits are obtained from the use of heavy loads while muscle hypertrophy can be equally achieved across a spectrum of loading ranges.

Honesty requires noting the wobble in the hypertrophy result. A separate meta-regression on the same data did show a trend favouring heavy loads (p = 0.10), and the analysis flagged five influential studies; removing each of the three most influential ones pushed that p-value to between 0.22 and 0.46 and shrank the difference further. The fair reading is not that load is provably irrelevant to growth, but that any advantage is too small for this evidence to pin down.

A much larger network, the same split

The strongest corroboration comes from a Bayesian network meta-analysis comparing twelve prescriptions — defined by load, sets and weekly frequency — against each other and against a non-exercising control.² Its strength network held 178 studies and 5,097 participants; its hypertrophy network, 119 studies and 3,364 participants.

Every prescription beat doing nothing for both outcomes. Beyond that, higher-load prescriptions (above 80% of 1RM) maximised strength, and — in the authors' words — all prescriptions comparably promoted muscle hypertrophy. Top-ranked for strength was higher-load, multi-set, thrice-weekly training, at a standardised mean difference of 1.60 (95% credible interval 1.38 to 1.82) against control; for hypertrophy, higher-load, multi-set, twice-weekly training, at 0.66 (0.47 to 0.85). Both are comparisons against no training at all, not heavy against light — the load effect lives in the ranking.

What does move hypertrophy

If load is not the lever for growth, something has to be. Two candidates have decent evidence.

The first is weekly volume. A meta-regression of 34 treatment groups from 15 studies found a graded relationship between weekly sets per muscle and change in muscle size (p = 0.002): each additional set was associated with an effect-size increase of 0.023, corresponding to 0.37% more growth.³ Comparing the higher- and lower-volume arm within each study gave an effect-size difference of 0.241, a 3.9% difference in percentage gain (p = 0.03). Splitting weekly sets into three bands — fewer than 5, 5 to 9, and 10 or more per muscle — produced only a trend (p = 0.074), a reminder that the dose–response is a slope, not a staircase with a magic step.

The second is how close to failure you stop, and here the literature is genuinely unsettled. A meta-analysis of 15 studies found a trivial advantage for training to set failure over stopping short, at an effect size of 0.19 (95% CI 0.00 to 0.37, p = 0.045); but when the analysis was restricted to the studies that used momentary muscular failure rather than looser definitions, the advantage disappeared (0.12, 95% CI −0.13 to 0.37, p = 0.343), as it did for higher versus moderate velocity-loss thresholds (0.08, −0.16 to 0.32, p = 0.529).⁴ They concluded the relationship is potentially non-linear.

A later set of meta-regressions treated proximity to failure as a continuous variable — repetitions in reserve — rather than a yes-or-no comparison, and recovered a cleaner picture.⁵ In every best-fitting strength model, the confidence interval around the slope for repetitions in reserve contained zero; in every best-fitting hypertrophy model, the slope was negative and its interval excluded zero. Their summary: Strength gains were similar across a wide range of RIR, while muscle hypertrophy improves as sets are terminated closer to failure. The same dissociation as load, from another direction — though the authors caution that repetitions in reserve were estimated from study descriptions rather than measured, and call their analysis exploratory.

Neither option is free

"Use whatever load you like" is not the same as "the two are interchangeable in practice". Light-load training to failure demands more repetitions, more time and more total work; Schoenfeld's group noted that it requires exercise volume (work) and time in excess of high-load training, suggesting high-load training may be more efficient.¹ The last repetitions of a set of thirty are unpleasant in a way a set of six is not; still, the trials in that review reported high adherence in both conditions, above 87% of sessions where attendance was recorded.

Heavy loading is widely assumed to carry its own bill: greater joint and connective-tissue stress, and less margin for a technical error under a maximal bar. On measured harms the review is reassuring but thin — only 2 of the 21 studies reported mild adverse effects, described as minor tendonitis and two minor injuries. That is a small denominator over short interventions, not a clean bill of health.

What these trials cannot tell you

Three limits matter. The trials are short: the load review required a minimum of six weeks, and most of its studies ran six to thirteen weeks, with one year-long exception. Nobody has run this comparison across the years over which a physique is actually built.

The participants are mostly beginners. The load review's own discussion concedes that its strength findings are primarily based on untrained subjects as only 3 studies investigated isotonic strength changes between conditions in those with RT experience, and a sub-analysis suggested the advantage of heavy loading was larger in trained participants. Sex is a weaker objection than it is often made: the network meta-analysis reported 45% women in its strength network and 47% in its hypertrophy network.²

Finally, "hypertrophy" here means images of particular muscles — magnetic resonance imaging, computed tomography or ultrasound at a handful of measured sites — not whole-body muscle mass, and the pooled estimate rested on 41 effect sizes from 10 studies. A site-specific measurement can miss growth elsewhere, or exaggerate it.

None of this needs software — a notebook does the job. Kettle records strength sessions alongside runs and rides on iPhone, and the data stays on the phone unless you give permission to write it to Apple Health.