This aims of this study were (I) to determine the velocity variable and regression model which best fit the load-velocity relationship during the free-weight prone bench pull exercise, (II) to compare the reliability of the velocity attained at each percentage of the one-repetition maximum (1RM) between different velocity variables and regression models, and (III) to compare the within- and between-subject variability of the velocity attained at each %1RM. Eighteen men (14 rowers and four weightlifters) performed an incremental test during the free-weight prone bench pull exercise in two different sessions. General and individual load-velocity relationships were modelled through three velocity variables (mean velocity [MV], mean propulsive velocity [MPV] and peak velocity [PV]) and two regression models (linear and second-order polynomial). The main findings revealed that (I) the general (Pearson’s correlation coefficient [r] range = 0.964–0.973) and individual (median r = 0.986 for MV, 0.989 for MPV, and 0.984 for PV) load-velocity relationships were highly linear, (II) the reliability of the velocity attained at each %1RM did not meaningfully differ between the velocity variables (coefficient of variation [CV] range = 2.55–7.61% for MV, 2.84–7.72% for MPV and 3.50–6.03% for PV) neither between the regression models (CV range = 2.55–7.72% and 2.73–5.25% for the linear and polynomial regressions, respectively), and (III) the within-subject variability of the velocity attained at each %1RM was lower than the between-subject variability for the light-moderate loads. No meaningful differences between the within- and between-subject CVs were observed for the MV of the 1RM trial (6.02% vs. 6.60%; CV ratio = 1.10), while the within-subject CV was lower for PV (6.36% vs. 7.56%; CV ratio = 1.19). These results suggest that the individual load-MV relationship should be determined with a linear regression model to obtain the most accurate prescription of the relative load during the free-weight prone bench pull exercise.