In this study, a statistical modeling approach using Stepwise Quadratic Regression was employed to predict the surface roughness (Ra) in the Powder-Mixed Electrical Discharge Machining (PMEDM) process of hardened 90CrSi steel. Silicon carbide (SiC) powder was added to the dielectric fluid to enhance machining performance. The effects of key process parameters, including powder concentration (Cp), pulse-on time (Ton), pulse-off time (Toff), peak current (IP), and servo voltage (SV), on Ra were analyzed. A response surface model with reduced complexity was developed using the stepwise selection method, resulting in a high level of predictive accuracy (R2 = 1.0000, adjusted R2 = 0.9999). The model was further used for constrained optimization to determine the optimal parameter combination that minimizes Ra while ensuring physical validity (Ra ≥ 1 µm). The findings confirm that the stepwise-enhanced quadratic regression model is a reliable and interpretable tool for modeling and optimizing surface roughness in PMEDM processes.