This study aimed to compare the reliability and validity of various prediction models based on load-velocity relationships for predicting the one-repetition maximum (1 RM) in Bulgarian split squat (BSS) exercises. Twenty-seven resistance-trained men participated in the study, completing both a 1-RM test and a progressive loading test to determine the 1 RM value, along with the mean velocity (MV) and peak velocity (PV) for each load. Load-velocity relationships were constructed using linear and binomial regression equations. The results revealed a strong correlation between different velocities and relative loads in BSS exercises (R² = 0.880–0.964), with the MV-based model slightly outperforming the PV-based model (R² difference of 0.06 and SEE difference of 0.02). Additionally, both MV and PV across all test loads demonstrated good reliability (ICC ≥ 0.801; CV ≤ 2.36%). Despite very strong correlations between the actual and predicted 1 RM across all models (R = 0.890–0.972), linear regression models and binomial regression models for the dominant leg consistently underestimated the actual 1 RM (p ≤ 0.035). Only the binomial regression models for the non-dominant leg accurately predicted the 1 RM (p ≥ 0.223). In conclusion, modeling using MV and binomial regression provides more reliable and accurate 1 RM predictions for BSS exercises. These findings confirm the utility of load-velocity relationships in predicting 1 RM for BSS exercises and suggest that practitioners should use MV and polynomial regression equations for predicting 1 RM in highly complex free-weight resistance exercises.