Purpose: To compare the accuracy of nine 1-repetition maximum (1RM) prediction methods during the paused and touch-and-go bench press exercises performed in a Smith machine. Method: A total of 86 men performed 2 identical sessions (incremental loading test until reaching the 1RM followed by a set to failure) in a randomized order during the paused and touch-and-go bench press exercises. Individualized load–velocity relationships were modeled by linear and polynomial regression models considering 4 loads (45%–60%–75%–90% of 1RM) (multiple-point methods) and considering only 2 loads (45%–90% of 1RM) by a linear regression (2-point method). Three minimal velocity thresholds were used: the general velocity of 0.17 m·s⁻¹ (general velocity of the 1RM [V_1RM]), the velocity obtained when lifting the 1RM load (individual V_1RM), and the velocity obtained during the last repetition of a set to failure. Results: The 1RM prediction methods were generally valid (range: r = .96–.99, standard error of the estimate = 2.8–4.9 kg or 4.6%–8.0% of 1RM). The multiple-point linear method (2.79 [2.29] kg) was more precise than the multiple-point polynomial method (3.54 [3.31] kg; P = .013), but no significant differences were observed when compared with the 2-point method (3.09 [2.66] kg, P = .136). The velocity of the last repetition of a set to failure (3.47 [2.97] kg) was significantly less precise than the individual V_1RM (2.91 [2.75] kg, P = .009) and general V_1RM (3.00 [2.65] kg, P = .010). Conclusions: Linear regression models and a general minimal velocity threshold of 0.17 m·s⁻¹ should be recommended to obtain a quick and precise estimation of the 1RM during the bench press exercise performed in a Smith machine.