Abstract Catalyst size, which determines surface area, is one of the major factors in catalytic performance. In this study, response surface methodology (RSM) and an adaptive neuro-fuzzy inference system (ANFIS) were applied to quantify the effects of physical characteristics of magnetite on Fenton-like oxidation efficiency of methylene blue. For this purpose, two magnetite samples (M and N) were used and characterized by XRD, BET surface area, particle size analyzer and FE-SEM. Central composite design (CCD) was applied to design the experiments, develop regression models, optimize and evaluate the individual and interactive effects of five independent variables: H2O2 and catalyst concentrations, pH, reaction time (numeric factors) and the type of catalyst (categorical factor). For each categorical factor, three quadratic models were developed regarding target responses: decolorization (YMB), COD (YCOD) and TOC (YTOC) removal efficiencies (%). The quadratic models were estimated by CCD and ANFIS methodologies. ANFIS was implemented using Matlab/Simulink and the performances were investigated. ANFIS models performed better for catalyst N compared to catalyst M, for color, COD and TOC separately. On contrary, it performed better for catalyst M compared to catalyst N, for combinations of color, COD and TOC. The obtained RMSE and R2 for the ANFIS networks show the effectiveness of catalyst N compared to catalyst M in Fenton oxidation process.