This study aimed to compare the within-session reliability and magnitude of velocity variables recorded against a range of submaximal loads during the bench press (BP) exercise performed in a Smith machine using different grip widths. Sixteen physically active men (BP one-repetition maximum [1RM] relative to body mass = 1.01 ± 0.19 kg·kg −1) were randomly tested on 4 sessions using a close grip width (100% of biacromial width), medium grip width (150% of biacromial width), wide grip width (200% of biacromial width), and self-selected grip width (176 ± 17% of biacromial width). Mean velocity (MV), maximum velocity (Vmax), and vertical displacement were recorded with a linear velocity transducer against the 35%1RM, 55%1RM, and 75%1RM. The main findings revealed that (I) the self-selected was the only grip width with an acceptable reliability for all loads and velocity variables (CV ≤ 7.56%; ICC ≥ 0.82), (II) the medium grip width provided the highest reliability for MV (CV ratio ≥ 1.20), while a comparable reliability was observed for Vmax using the close, medium and self-selected grip widths (CV ratio ≤ 1.08), (III) the Vmax showed the highest reliability for all grip widths (CV ratio = 1.68), and (IV) the MV and vertical displacement of the barbell were generally higher for narrow grip widths (close and medium) compared to the wide and self-selected grip widths, while no significant differences between the grip widths were observed for Vmax (p >.05). Taken together, we recommend the assessment of Vmax using a self-selected grip width during the routine testing of BP performance against submaximal loads.