This study examined the reliability and validity of three methods of estimating the one-repetition maximum (1RM) during the free-weight prone bench pull exercise. Twenty-six men (22 rowers and four weightlifters) performed an incremental loading test until reaching their 1RM, followed by a set of repetitions-to-failure. Eighteen participants were re-tested to conduct the reliability analysis. The 1RM was estimated through the lifts-to-failure equations proposed by Lombardi and O'Connor, general load-velocity (L-V) relationships proposed by Sánchez-Medina and Loturco and the individual L-V relationships modelled using four (multiple-point method) or only two loads (two-point method). The direct method provided the highest reliability (coefficient of variation [CV] = 2.45% and intraclass correlation coefficient [ICC] = 0.97), followed by the Lombardi's equation (CV = 3.44% and ICC = 0.94), and no meaningful differences were observed between the remaining methods (CV range = 4.95-6.89% and ICC range = 0.81-0.91). The lifts-to-failure equations overestimated the 1RM (3.43-4.08%), the general L-V relationship proposed by Sánchez-Medina underestimated the 1RM (-3.77%), and no significant differences were observed for the remaining prediction methods (-0.40-0.86%). The individual L-V relationship could be recommended as the most accurate method for predicting the 1RM during the free-weight prone bench pull exercise.