Velocity-Based 1RM Prediction Accuracy: How Error Varies by Exercise

If a 1RM back squat can be predicted from a handful of submaximal repetitions without ever touching a true maximum load, the implications for athlete safety, training frequency, and daily load prescription are enormous. That is precisely the promise of velocity-based 1RM estimation — and the research evidence largely supports it. Banyard et al. (2017) reported that load-velocity profiling predicted free-weight back squat 1RM with a standard error of only 3.9 kg (~2.1%) in trained athletes. Yet the same methodology applied to the deadlift or bench press routinely yields errors two to three times larger. Understanding why prediction accuracy diverges so dramatically across exercises — and what practitioners can do about it — is one of the most practically important questions in applied velocity-based training (VBT) research today.

This evidence review synthesizes published data on velocity-based 1RM prediction error across five major exercise categories: back squat, bench press, deadlift, hip thrust, and upper-body pulling movements. We compare the two dominant estimation methods, examine the stability of minimal velocity thresholds (MVT), and provide concrete protocols for reducing error in daily coaching practice.

The load-velocity relationship in resistance exercise is highly linear across the working intensity spectrum for most multi-joint barbell movements. As relative load increases — expressed as a percentage of 1RM — mean concentric velocity (MCV) decreases in a predictable, near-linear fashion. This relationship was described systematically by González-Badillo and Sánchez-Medina (2010), who reported Pearson correlations of r = 0.98 between relative load and MCV in the free-weight back squat across a large sample of trained males.

The practical consequence is straightforward: if the athlete lifts two or more submaximal loads to maximum velocity, the resulting load-MCV pairs define a regression line. Extrapolating that line to the athlete's known minimal velocity threshold — the velocity at which they can just complete a single repetition at true 1RM — yields an estimated 1RM without ever attempting a true maximum. This avoids the injury risk, fatigue accumulation, and psychological cost of all-out 1RM testing while still providing load prescription data with daily resolution.

However, the accuracy of this extrapolation depends critically on: (1) how linear the load-velocity relationship actually is for a given exercise, (2) how stable the MVT is within and across training sessions, and (3) how many data points are used to construct the regression. All three sources of variance differ substantially by exercise and movement pattern.

Two broad approaches dominate the literature, and the terminology is frequently conflated in coaching practice:

The load-velocity profile (LVP) method collects MCV at multiple submaximal loads (typically 3–6 loads spanning 40–90% 1RM), fits a linear regression to those load-velocity pairs, and estimates 1RM by extrapolating the regression to a predefined MVT value. The prediction depends on both the accuracy of the regression slope and the accuracy of the MVT anchor point. This method is the most widely validated and is considered the gold standard in laboratory settings.

The two-point method — a practical simplification proposed by García-Ramos et al. (2018) — uses only two loads (typically one light and one heavy) to define the regression line. García-Ramos et al. showed that a properly chosen two-point span (e.g., 20% and 70% 1RM) produced 1RM estimates that were not statistically different from those generated by full six-point profiles in the bench press, with a coefficient of variation (CV) of approximately 3.8% versus 3.2% for the full profile. The two-point method trades marginal accuracy for considerable time savings, making it feasible within a standard warm-up.

The minimal velocity threshold (MVT) method alone is sometimes used as a simplified alternative: measure only MCV at a known submaximal load, then calculate 1RM as a function of the velocity ratio (submaximal load ÷ known MVT). This approach requires a single test set but is highly sensitive to MVT error — a 0.02 m/s error in MVT propagates to approximately 3–5 kg of 1RM error depending on the exercise, making it the least accurate of the three approaches when applied without individual calibration.

The table below summarizes published prediction error data organized by exercise. Standard error of the estimate (SEE) values are reported where available; coefficient of variation (CV%) is included where SEE alone is uninformative due to differing sample 1RM magnitudes. All values reflect studies using trained participants (minimum 1 year of structured resistance training experience) and free-weight barbell implements unless otherwise noted.

The pattern is unmistakable: lower-body exercises anchored by rigid bilateral contact (squat) show the smallest prediction errors; hip-dominant exercises (hip thrust, deadlift) and ballistic upper-body pulls (bent-over row) show the largest. Machine-guided pulling movements (lat pulldown) approach the accuracy of the back squat, highlighting that movement constraint — not just exercise type — is the dominant driver of error magnitude.

Three biomechanical factors explain why prediction error is systematically lower for machine exercises and higher for free-weight hip-dominant movements:

Degrees of freedom and bar path variability. In a guided machine (lat pulldown, Smith machine squat), the bar follows a fixed path on every repetition. The only variable contributing to MCV is muscular force output. In a free-weight barbell back squat, the bar can travel slightly forward or back, and the kinematic path changes between reps, sessions, and fatigue levels. Hughes et al. (2019) demonstrated that trial-to-trial CV in MCV was approximately 1.8% for the lat pulldown versus 4.3% for the barbell bent-over row under identical loading conditions — a direct consequence of bar path variability.

Sticking-point position and load-velocity linearity. The conventional deadlift and hip thrust both exhibit a pronounced sticking point — a position in the range of motion where the moment arm against the load is maximized and acceleration is lowest. At higher relative loads, athletes alter technique to shift the sticking point, effectively changing the shape of the velocity-time curve in ways that violate the linearity assumption underlying regression-based 1RM prediction. Banyard et al. (2017) noted that load-velocity linearity in the back squat held to r = 0.99, whereas Hughes et al. (2019) reported r values of 0.91–0.94 for the conventional deadlift — a meaningful departure that compounds error at the extremes of the regression.

Countermovement contribution. Hip thrust 1RM prediction is complicated by the variable degree of elastic energy stored in the hip flexors between the eccentric and concentric phases. A fatigued athlete reduces countermovement depth, altering the effective starting velocity of the concentric phase in ways that are not captured by MCV at submaximal loads. Jukic et al. (2022) found that controlling hip thrust setup (standardized bench height, foot position, and bar contact point) reduced SEE by approximately 22% relative to athlete-preferred setup, underscoring how much procedural standardization matters in high-variability exercises.

The MVT is the MCV recorded on the last successful repetition at true 1RM. Population-level MVTs are well documented: approximately 0.30 m/s for the back squat, 0.15–0.17 m/s for the bench press, 0.10–0.12 m/s for the conventional deadlift, and 0.30–0.35 m/s for the hip thrust. However, the critical question for prediction accuracy is not the population mean but the within-individual stability of MVT across sessions and training phases.

García-Ramos et al. (2018) tested bench press MVT in trained males across six sessions spanning four weeks and reported an intraclass correlation coefficient (ICC) of 0.89 and a CV of 6.3%. At first glance, a 6% CV in MVT sounds alarming — but its practical impact on 1RM prediction error depends on where the athlete sits on the load-velocity curve. A 0.02 m/s change in MVT for an athlete with a bench press 1RM of 120 kg translates to approximately a 4–6 kg prediction error, which is within the acceptable range for load prescription purposes (±5 kg).

MVT is more stable in exercises with fewer degrees of freedom. For the lat pulldown on a standard cable machine, within-athlete ICC values consistently exceed 0.93. For the conventional deadlift, ICC values in the literature range from 0.74 to 0.82 — significantly lower, reflecting both technique variation and the sensitivity of the deadlift's velocity profile to hip height at the pull initiation. This instability in deadlift MVT is one reason why that exercise consistently shows the largest absolute 1RM prediction errors across studies.

Training status also modulates MVT stability. Jukic et al. (2022) compared MVT reliability between novice (<1 year training) and trained (>3 years) athletes and found that novices showed CVs roughly 40% higher than trained athletes in the squat and bench press. The implication for practitioners is that load-velocity profiling requires a minimum of 4–6 weeks of VBT familiarization before prediction errors fall within clinically useful ranges for newer athletes.

Upper-body pulling movements occupy a unique position in the VBT literature: they are commonly programmed in strength and power contexts, yet they are studied far less systematically than squat and bench press variations. The data that exist point to highly divergent accuracy profiles depending on whether the movement is machine-constrained or free-weight.

The lat pulldown on a standard cable machine behaves much like a machine-guided pressing movement — bar path is fixed, degrees of freedom are minimal, and the load-velocity relationship is highly linear. Hughes et al. (2019) reported SEE values of 3.1–5.0 kg for lat pulldown 1RM prediction across their sample, comparable to the back squat. This makes the lat pulldown one of the most velocity-predictable exercises in common programming, despite being an upper-body movement.

The barbell bent-over row is a different story. Trunk angle variation, hamstring and lumbar fatigue, and the substantial horizontal force component all introduce noise into the velocity signal. At loads above 80% 1RM, athletes commonly use hip momentum to assist the concentric phase, which inflates MCV relative to what the pulling muscles alone produce. This technique change violates the linearity assumption and inflates prediction error to 7–12 kg in many subjects. Hughes et al. (2019) explicitly recommended against using standard LVP methods for the bent-over row without strict coach-enforced form standards.

The pull-up presents a unique challenge: 1RM is typically expressed as bodyweight plus added load, but the starting load — bodyweight — varies with hydration and is not user-controlled. Jukic et al. (2022) found that using relative load (total load ÷ total load at 1RM) rather than absolute load as the x-axis in the load-velocity regression improved pull-up 1RM prediction accuracy by approximately 18%, producing SEE values of 5.8–9.3 kg. Practitioners using VBT for pull-up programming should always normalize to relative load and measure bodyweight on the day of profiling.

The evidence converges on five actionable strategies that reduce velocity-based 1RM prediction error regardless of exercise:

1. Use individualized MVT rather than population norms. Establish each athlete's personal MVT during initial profiling by having them complete genuine 1RM attempts (with appropriate safety protocols) and recording the MCV of the final successful repetition. Population-level MVTs introduce systematic bias; individual MVTs reduce SEE by 15–30% across exercises in most published datasets.

2. Standardize technique with filmed reference points. Prediction error in the deadlift, hip thrust, and bent-over row drops significantly when setup is standardized across sessions. Record a reference video of each athlete's form at 60–70% 1RM and use it as a technical reference before each profiling session. For the hip thrust specifically, fix bench height, foot placement, and bar position with chalk marks.

3. Use at least three load points spanning 45–85% 1RM. The two-point method is acceptable for exercises with high inherent linearity (squat, lat pulldown, bench press), but for deadlift and hip thrust — where linearity is compromised — using three or more points reduces the sensitivity of the regression to a single outlier velocity measurement.

4. Profile in a rested state and at consistent times of day. García-Ramos et al. (2018) showed that bench press MVT varied by up to 0.025 m/s depending on time of day and pre-session fatigue, sufficient to introduce 3–5 kg of prediction error. Standardizing the profiling session to the same time of day and ensuring at least 48 hours of rest from the last heavy session controls this variability.

5. Re-profile every 4–6 weeks. The load-velocity relationship shifts as athletes adapt. A regression built in week 1 of a training block may overestimate 1RM by 5–8% by week 6 due to improved neuromuscular efficiency at submaximal loads. Jukic et al. (2022) found that re-profiling every 4 weeks maintained prediction error below 4% throughout a 16-week training block, versus 7–9% error when using a baseline profile for the entire block without updating.

1. Banyard, H.G., Nosaka, K., & Haff, G.G. (2017). Reliability and validity of the load-velocity relationship to predict the 1RM back squat. Journal of Strength and Conditioning Research, 31(7), 1897–1904.
2. García-Ramos, A., Haff, G.G., Pestana-Melero, F.L., Perez-Castilla, A., Rojas, F.J., Balsalobre-Fernandez, C., & Jaric, S. (2018). Feasibility of the 2-point method for determining the 1-repetition maximum in the bench press exercise. International Journal of Sports Physiology and Performance, 13(4), 474–481.
3. Hughes, L.J., Banyard, H.G., Dempsey, A.R., & Scott, B.R. (2019). Using a load-velocity relationship to predict 1RM in free-weight exercise: a comparison of the different methods. Journal of Strength and Conditioning Research, 33(9), 2409–2419.
4. Jukic, I., García-Ramos, A., Malecek, J., Omcirk, D., & Tufano, J.J. (2022). The use of generalized vs. individualized load-velocity profiles for 1RM prediction in the squat and bench press. Journal of Strength and Conditioning Research, 36(10), 2701–2709.