Purpose: This study assessed the reliability and load–velocity profiles of 3 different landmine-punch-throw variations (seated without trunk rotation, seated with trunk rotation, and standing whole body) with different loads (20, 22.5, and 25.0 kg), all with the dominant hand and nondominant hand. Methods: In a quasi-randomized order, 14 boxers (24.1 [4.3] y, 72.6 [10.1] kg) performed 3 repetitions of each variation with their dominant hand and their nondominant hand, with maximal effort and 3 minutes of interset rest. Peak velocity was measured via the GymAware Power Tool (Kinetic Performance Technologies). The interclass correlation coefficients and their 95% CIs were used to determine the intrasession reliability of each variation × load × hand combination. Additionally, a 2 (hand) × 3 (variation) repeated-measures analysis of variance assessed the load–velocity profile slope, and a 3 (variation) × 2 (hand) × 3 (load) repeated-measures analysis of variance assessed the peak velocity of each variation. Results: Most variations were highly reliable (intraclass correlation coefficient > .91), with the nondominant hand being as reliable or more reliable than the dominant hand. Very strong linear relationships were observed for the group average for each variation (R2 ≥ .96). However, there was no variation × hand interaction for the slope, and there was no main effect for variation or hand. Additionally, there was no interaction for the peak velocity, but there were main effects for variation, hand, and load (P < .01). Conclusion: Each variation was reliable and can be used to create upper-body ballistic unilateral load–velocity profiles. However, as with other research on load–velocity profile, individual data allowed for more accurate profiling than group average data.