Competitive weightlifting, as an Olympic sport, is becoming increasingly popular worldwide (Storey and Smith, 2012). In competition, athletes aim to lift the heaviest weight in snatch, clean & jerk, or their combined total within a bodyweight category (Sandau et al., 2022b). The heaviest successful lift is also defined as the one-repetition maximum (1RM), which is widely regarded as the gold standard for assessing maximal dynamic strength (Todd et al., 2012). Coaches typically choose a near-maximal load for the opener (i.e., first attempt in the weightlifting competition) based on previous training performance, competition-day readiness, and rival attempts, a decision that is largely subjective and based on the coaches’ experience. An opener set too heavy risks a failed attempt, while setting the opener too light may result in a missed opportunity to challenge a heavier load. This decision is challenging, as 1RM performance can be influenced by various factors such as mental stress and weight cutting (Kwan and Helms, 2022; Storey et al., 2016; Williams et al., 2017). Therefore, accurately predicting 1RM is crucial to optimizing opener selection in weightlifting competitions.

Velocity-based methods have gained popularity as an alternative to traditional 1RM testing, particularly in common resistance exercises such as the bench press (Janićijević et al., 2021), back squat (Chen et al., 2025a), and deadlift (Janićijević et al., 2024). Among these methods, the most accurate approach for estimating 1RM involves assessing the individual load–velocity relationship and then applying a linear regression equation to estimate the 1RM as the load corresponding to a predefined minimal velocity threshold (MVT) (García-Ramos, 2023b). For this purpose, either the lifting velocity associated with a previously assessed individual 1RM (i.e., actual MVT) or the average 1RM velocity reported across subjects in prior studies (i.e., general MVT) has been used (García-Ramos, 2025). Moreover, García-Ramos (2023a) recently proposed the concept of an optimal MVT, defined as the velocity threshold that would eliminate discrepancies between actual and predicted 1RM values within the same testing session. Existing research has shown that optimal MVT provides more accurate 1RM predictions than traditional (actual and general) MVTs in both the bench press (García-Ramos, 2023a) and back squat (Chen et al., 2025a; Fitas et al., 2024). When applied to weightlifting exercises, Haff et al. (2020) reported that the actual MVT failed to accurately predict the power clean 1RM in weightlifting enthusiasts. However, no study has examined whether using the optimal MVT could improve the accuracy of 1RM estimation in weightlifting exercises. If confirmed, this approach could provide weightlifting coaches with objective data to guide attempt selection in competition.

The accuracy of 1RM prediction may depend on the reliability of the MVT used in the estimation process (Chen et al., 2025a). Thompson et al. (2021) found that the actual MVT of the power clean demonstrated high reliability in competitive weightlifters, although they did not investigate its accuracy for 1RM prediction. Previous studies have shown that the optimal MVT demonstrates higher reliability than the actual MVT (Chen et al., 2025a; Fitas et al., 2024; García-Ramos, 2023a), which may contribute to improved accuracy in 1RM prediction. However, existing research on the optimal MVT just focused on traditional exercises, (Chen et al., 2025a; Fitas et al., 2024; García-Ramos, 2023a; Janićijević et al., 2024; Miras-Moreno and García-Ramos, 2024). Therefore, whether the optimal MVT can demonstrate higher reliability than the actual MVT in weightlifting exercises, and whether it can provide more accurate 1RM predictions, warrants further investigation. On the other hand, the reliability of the individual %1RM-velocity relationship has been extensively investigated in traditional exercises (Banyard et al., 2018; Chen et al., 2025b; Thompson et al., 2021). Since competitive weightlifters routinely test their 1RM as a direct measure of performance and exhibit superior movement proficiency (Sandau and Granacher, 2022a), testing single loads and applying the individual %1RM-velocity relationship from a previous session may offer comparable - or even greater - predictive accuracy than relying on actual or optimal MVTs. Although the individual %1RM-velocity relationship has long been proposed as a method to regulate training loads (Thompson et al., 2025; Weakley et al., 2021), its accuracy in predicting 1RM has yet to be validated in both weightlifting and conventional exercises.

To address the gaps in previous research, this study had two main objectives. The first objective was to examine the between-sessions reliability of the actual MVT, optimal MVT, and peak velocity (PV) associated with 50%, 70%, 80%, and 90% 1RM of the snatch. We hypothesized that the between-sessions reliability of the optimal MVT and PVs associated with submaximal loads would be higher than that of the actual MVT (Chen et al., 2025a). The second objective was to evaluate the accuracy of different velocity-based methods for predicting snatch 1RM in a follow-up session: (i) using the load-PV relationship from four loads combined with either the actual or optimal MVT, and (ii) estimating 1RM from PV recorded at single loads (50%, 70%, 80%, or 90% 1RM) using the individual %1RM-PV relationship established previously. We hypothesized that the individual %1RM-PV relationship using heavier loads (e.g., 80% and 90% 1RM) and the optimal MVT method would provide more accurate 1RM predictions than the actual MVT method and the individual %1RM-PV relationship using lower loads (e.g., 50% and 70% 1RM) (Chen et al., 2025a; Thompson et al., 2021).

A repeated-measures design was used to assess the reliability and accuracy of different velocity-based methods for predicting snatch 1RM. Participants completed two testing sessions separated by one week. Each session began with lifts at 50%, 70%, 80%, and 90% of their best snatch performance from the previous 30 days, followed by progresssive load increases until their 1RM for that day was determined. Between-sessions reliability was assessed for actual MVT, optimal MVT, and PV at 50%, 70%, 80%, and 90% 1RM. To evaluate the accuracy of 1RM prediction, actual MVT, optimal MVT, and the individual %1RM-PV relationship - all derived from session one - were used to predict 1RM in session two. All testing was conducted at 2:30 p.m. during participants’ regular training sessions under consistent environmental conditions (~20°C, ~65% humidity).

Considering an effect size of f = 0.25, an ɑ error probability of 0.05, a power of 0.80, 1 group, 6 measurements, and assuming a correlation among repeated measures of 0.75, the sample size calculation indicated that at least 10 participants are required for detecting the postulated effects (G*Power software, version 3.1.9.6) (Miras-Moreno et al., 2024). Fourteen male competitive adolescent weightlifters (age: 15.9 ± 0.9 years; body height: 170.0 ± 6.2 cm; body mass: 67.6 ± 9.3 kg; best snatch record from the past 30 days: 103.4 ± 15.1 kg) were recruited for this study. All participants had received professional and systematic weightlifting training (5.1 ± 0.9 years) at a local weightlifting school and were proficient in weightlifting techniques. No physical limitations or musculoskeletal injuries that could compromise weightlifting performance were reported. Testing was conducted only when participants, assessed on a readiness scale from 1 (poor) to 5 (well), reported a relaxed state (> 3). No participant required rescheduling of testing due to insufficient readiness. Participants and their parents and coaches were informed of the procedures and signed a written informed consent form before initiating the study. The study protocol adhered to the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board (IRB approval: JMU202411088).

Each testing session began with a 10-minute warm-up, which was part of the participants’ daily pre-training routine and included dynamic stretching for the upper and lower limbs, as well as weightlifting technique drills with an empty barbell. After a 3-minute rest, participants commenced the incremental load testing protocol. The protocol consisted of three lifts at 50% (51.5 ± 7.6 kg), two lifts at 70% (72.1 ± 10.6 kg), one lift at 80% (82.7 ± 12.1 kg) and one lift at 90% (93.1 ± 13.7 kg) of their previous snatch 1RM record. The selection of testing loads was based on the recommendations of García-Ramos (2023c), who suggested using a lightest load within 40-50% 1RM and a heaviest load around 90% 1RM, complemented with appropriate intermediate loads to ensure a representative load-velocity relationship. The load was then progressively increased in 1 to 5 kg increments until participants achieved their actual 1RM, with a maximum of 5 attempts allowed. Most participants reached their 1RM within 2-4 attempts. Consistent with their regular training routine, participants rested approximately 1 minute between repetitions and 3 minutes between load increases. The fastest PV for each load was included in the data analysis (Haff et al., 2020). It is important to note that mean velocity (MV) is not suitable for monitoring the snatch, as the exercise involves two distinct ascending phases. The execution standards for the snatch in this study were consistent with the International Weightlifting Federation rules.

An Olympic barbell and weight plates (ZKC, Cangzhou, China) were used in all testing sessions, coupled with a linear position transducer (GymAware RS, Kinetic Performance Technology, Canberra, Australia; sampling rate = 100 Hz), which has demonstrated to be valid for monitoring lifting velocity in weightlifting exercises (Thompson et al., 2021). The linear position transducer was positioned on the ground to the left of the participants’ feet, with the Velcro strap attached to the left end of the barbell, ensuring that the athlete’s technical movements were not obstructed.

For each load, only the fastest repetition was considered for analysis. The actual MVT was defined as the PV recorded during the 1RM lift of that session. The absolute loads corresponding to 50%, 70%, 80%, and 90% of each participant’s previous best snatch and their associated PVs were used to create a linear regression model, determining the load-axis intercept and slope of the load-PV relationship. The optimal MVT was then calculated by solving for velocity using the actual 1RM load in the load-PV equation: optimal MVT = (actual 1RM - load intercept)/slope (Chen et al., 2025a; García-Ramos, 2023a; Miras-Moreno and García-Ramos 2024). Similarly, the individual %1RM-PV relationship was established using the relative load to the actual 1RM obtained in that session.

In the second session, a new load-PV relationship was developed using the same procedure. The 1RM for the second session was predicted by applying the MVTs from the first session. Additionally, each absolute load (50%, 70%, 80% and 90% 1RM) and corresponding PV from the second session were individually applied to the individual %1RM-PV relationship from the first session to estimate the 1RM of the second session (Sánchez-Medina et al., 2017). Specifically, the PV corresponding to an absolute testing load from the second session was substituted into the %1RM–PV relationship equation developed in the first session to determine the estimated %1RM for that absolute load. The 1RM was then calculated using the equation: estimated 1RM = absolute load / estimated %1RM. For example, if 60 kg was lifted at a PV of 2.6 m/s, and 2.6 m/s corresponded to 60% 1RM from session 1, then the estimated 1RM would be calculated as 60 / 0.6 = 100 kg.

Descriptive data were assessed for normal distribution using Shapiro-Wilk test and are presented as means and SD (Normal distribution) or medians and IQR (Non-normal distribution). Paired samples t tests, Cohen’s d effect size (ES), intraclass correlation coefficient (ICC, model 3.1), and within-subjects coefficient of variation (CV) were used to investigate the between-session reliability of the basic data (actual 1RM, PVs corresponding to different testing loads, actual MVT, and optimal MVT) (Hopkins, 2015). The goodness-of-fit of both the average and individualised load-PV and %1RM-PV relationships were assessed through the coefficient of determination (r²). A Friedman test with Dunn-Bonferroni post hoc corrections was applied to absolute errors between the actual and predicted 1RMs. The validity of predicted 1RMs with respect to the actual 1RM was examined through the Bland-Altman analysis. The magnitude of ES was interpreted as trivial (< 0.20), small (0.20-0.59), moderate (0.60-1.19), large (1.20-2.00), and very large (> 2.00) (Hopkins et al., 2009). Acceptable reliability was determined as an ICC > 0.7 and CV < 10% (Miras-Moreno et al., 2023). Proportional bias or heteroscedasticity was considered present when a significant linear regression was observed between the mean of the predicted and actual 1RM and the raw error or absolute error, respectively (Bland et al., 1999). Statistical analyses were performed using SPSS software (version 27.0, SPSS Ins., Chicago). Alpha was set at 0.05. Additionally, an exploratory supplementary analysis was conducted after completing the main analyses, in which individual data points with large prediction errors (>10 kg; n = 4) were removed to examine the CVs for the actual and optimal MVTs. The results of this analysis are presented in the Discussion section.

Non-significant (p ≥ 0.219) and trivial to small (ES ≤ 0.23) differences were observed for basic variables between sessions (Table 1). The snatch 1RM demonstrated high between-sessions reliability (ICC = 0.98, CV = 2.16%). Notably, all PVs and MVTs exhibited acceptable between-sessions reliability (ICC = 0.76-0.90, CV = 1.82-3.31%).

Figure 1 illustrates the average load-PV and %1RM-PV relationships for both sessions. The individualized load-PV and %1RM-PV relationships, modeled using linear regression, exhibited a near perfect goodness-of-fit across the first (r² = 0.968 ± 0.027 [range from 0.915 to 0.999]) and second (r² = 0.960 ± 0.052 [range from 0.857 to 1.000]) sessions.

Figure 2 illustrates significant differences in 1RM prediction accuracy (Q = 13.5, p = 0.019). Post hoc comparisons showed that the individualized %1RM-V relationship using 80% 1RM was more accurate than using 50% 1RM (p = 0.018), while other methods showed no differences (p ≥ 0.129).

No systematic bias was detected for any method (p = 0.111-0.838, ES = -0.03 to 0.32); however, all methods showed large random errors (11.9-29.1 kg). Proportional bias was found in the 1RM predictions using actual MVT, optimal MVT, 80% 1RM, and 90% 1RM (p = 0.002-0.018), while heteroscedasticity was observed in the predictions using actual MVT and 90% 1RM (p = 0.022-0.026) (Figure 3, Table 2).

This study assessed the reliability and accuracy of different velocity-based methods for predicting snatch 1RM in adolescent competitive weightlifters. The main findings suggest that the actual MVT, optimal MVT, and PVs at submaximal loads demonstrated acceptable between-sessions reliability. Predicted 1RMs showed no systematic error from actual 1RM. Absolute errors were below 5 kg when using actual MVT, optimal MVT, and the %1RM-PV relationship using heavier loads (80% and 90% 1RM). However, these methods exhibited proportional bias, driven by a tendency to overestimate in stronger individuals and underestimate in weaker ones. Furthermore, heteroscedasticity was observed for the actual MVT and 90% 1RM methods, indicating that prediction errors tended to increase as athletes’ performance increased.

Previous studies have reported that the optimal MVT using MV is more reliable than the actual MVT in bench press (ICC = 0.56-0.73 vs. 0.34) (García-Ramos, 2023a) and back squat (CV = 7.1-9.3% vs. 16.6-17.0%) (Chen et al., 2025a). However, our results did not confirm this for the snatch, as the optimal MVT (CV = 3.31%) showed similar reliability compared to the actual MVT (CV = 2.61%) using PVs. The differences in reliability between both types of MVT were reduced after excluding from the analysis individuals with large prediction errors (> 10 kg) (CV = 1.97% for optimal MVT and 2.30% for actual MVT). Notably, PVs attained at single submaximal loads always demonstrated acceptable reliability (ICC = 0.76-0.90, CV = 1.82-3.21%). Unlike traditional exercises, the transition from the “2^nd pull” to “catch” in the snatch relies heavily on inertia, making success at 1RM dependent on a specific PV threshold among other factors. This aligns with Thompson et al. (Thompson et al., 2021), who found greater actual MVT reliability in power clean than in squat among competitive weightlifters (CV = 8.1% vs. 27.8%). Although weightlifting exercises are typically considered ballistic movements, they differ from exercises like the bench press throw or squat jump, where maximal effort can be exerted even with light loads (García-Ramos, 2023c). In weightlifting, catching the barbell limits true maximal intent at lower loads (Sandau et al., 2021). In this study, 50% 1RM was selected as the initial load to establish the load-PV relationship, with its PV showing high reliability (ICC = 0.76, CV = 3.21%), and near-perfect linearity of the load-PV relationship (r² = 0.968 ± 0.027 in session one, 0.960 ± 0.052 in session two), supporting its suitability for determining the optimal MVT in the snatch. However, as shown in the lower panel of Figure 1, the range of PV values observed around 50% 1RM is highly variable between sessions, suggesting that the %1RM-PV relationship may be more reliable when using heavier loads as the lightest load.

Contrary to our hypothesis, actual and optimal MVTs showed comparable accuracy in predicting snatch 1RM, and given their low absolute errors, either could be used for estimation. When using the %1RM-PV relationship, absolute errors increased as the experimental points were farther from 1RM, with errors of 9.9 kg at 50% 1RM and 6.2 kg at 70% 1RM, suggesting these loads are unsuitable for snatch 1RM prediction. This aligns with García-Ramos’s perspective that prediction errors increase when experimental points are farther from the target value (García-Ramos, 2023c). Interestingly, predictions using the 90% 1RM were not more accurate than using the 80% 1RM, differing from traditional exercises where accuracy improves with heavier testing loads (García-Ramos, 2023c). A likely explanation is that although 90% 1RM is closer to 1RM, it exhibited lower reliability compared to the 80% 1RM (ICC = 0.80 vs. 0.90; CV = 2.73% vs. 1.82%, respectively). Given the small absolute errors of the actual MVT, the optimal MVT and the %1RM-PV relationship using the 80 to 90% 1RM, any of these methods could serve as objective tools to assist coaches in selecting competition attempts after warm-up. However, while this approach introduces objective data to weightlifting, final decisions should not rely solely on velocity-based estimates. Snatch performance is influenced by additional factors such as barbell horizontal displacement, drop height, and the lifter’s velocity during the turnover phase (Chavda et al., 2021; Liu et al., 2018; Mastalerz et al., 2019; Nagao et al., 2019; Nagao et al., 2023).

Although the raw and absolute errors of 1RM predictions were within acceptable ranges for all methods except the %1RM-PV relationship using 50% and 70% 1RM, the presence of proportional bias and heteroscedasticity poses challenges for practical application. Competitive weightlifting require extremely high accuracy as performance differences among athletes in the same weight category are often minimal (Yan et al., 2024). This study found that the accuracy of velocity-based methods for predicting snatch 1RM in youth competitive weightlifters is questionable due to the presence of proportional bias - characterized by a tendency to overestimate 1RM in stronger athletes and underestimate it in weaker ones - when using the actual MVT, optimal MVT, 80% 1RM, and 90% 1RM methods. Additionally, heteroscedasticity was observed for the actual MVT and 90% 1RM methods, indicating that prediction errors tended to increase with athletes’ performance levels. Given the frequent use of 1RM testing in weightlifters’ training routines, determining competition attempts could be informed by a combination of the most recent 1RM performance, real-time subjective measures (e.g., readiness questionnaires), and real-time objective measures (e.g., changes in lifting velocity). Nevertheless, the feasibility and practical utility of this novel combined approach - incorporating velocity-based metrics - should be validated by future research in the snatch and other weightlifting exercises. Future research could compare the success rates of attempts selected using subjective, previous 1RM-based, and velocity-based methods, providing more practical guidance for coaches.

Although this study provides novel and practical insights for weightlifting competitors, it is not without limitations. First, all tests were conducted during the off-season. In competition, factors such as mental stress and weight reduction may influence lifting velocity, warranting further investigation. Second, the study population was limited to adolescent male athletes. Previous research suggests that females exhibit lower lifting velocities at lighter relative loads compared to males, while velocities at heavier loads are similar (Nieto-Acevedo et al., 2023). This difference could influence the slope of the load-PV relationship and, in turn, the optimal MVT value. Third, the load-PV and %1RM-PV relationships were established under progressive fatigue, as rest intervals (1 minute between repetitions and 3 minutes between loads) may not have allowed full recovery. While consistent with the athletes’ training routines, this could have affected lifting velocity and model accuracy, making the results more reflective of training rather than competition 1RM performance.

Our findings revealed that the actual MVT, optimal MVT, and the individual %1RM-PV relationship using approximately the 80% and 90% 1RM may accurately predict snatch 1RM. Among these methods, the %1RM-PV relationship using 80% 1RM may be the most practical for regular monitoring, as it requires lighter loads and less testing time. However, large random errors, proportional bias, and heteroscedasticity limited the ability of these methods to accurately predict performance across all athletes. Nevertheless, current findings suggest that recording PV during warm-up sets prior to competition could serve as a complementary variable to support and refine opener selection in weightlifting competitions.