This study aimed to compare the between-session reliability of the load–velocity relationship between (1) linear versus polynomial regression models, (2) concentric-only versus eccentric–concentric bench press variants, as well as (3) the within-participants versus the between-participants variability of the velocity attained at each percentage of the 1-repetition maximum. The load–velocity relationship of 30 men (age: 21.2 [3.8] y; height: 1.78 [0.07] m, body mass: 72.3 [7.3] kg; bench press 1-repetition maximum: 78.8 [13.2] kg) were evaluated by means of linear and polynomial regression models in the concentric-only and eccentric–concentric bench press variants in a Smith machine. Two sessions were performed with each bench press variant. The main findings were: (1) first-order polynomials (coefficient of variation: 4.39%–4.70%) provided the load–velocity relationship with higher reliability than the second-order polynomials (coefficient of variation: 4.68%–5.04%); (2) the reliability of the load–velocity relationship did not differ between the concentric-only and eccentric–concentric bench press variants; and (3) the within-participants variability of the velocity attained at each percentage of the 1-repetition maximum was markedly lower than the between-participants variability. Taken together, these results highlight that, regardless of the bench press variant considered, the individual determination of the load–velocity relationship by a linear regression model could be recommended to monitor and prescribe the relative load in the Smith machine bench press exercise.