The homolysis of peracetic acid (PAA) as a relevant source of free radicals (e.g., *OH) was studied in detail. Radicals formed as a result of chain radical reactions were detected with electron spin resonance and nuclear magnetic resonance spin trapping techniques and subsequently identified by means of the simulation-based fitting approach. The reaction mechanism, where a hydroxyl radical was a primary product of O-O bond rupture of PAA, was established with a complete assessment of relevant reaction thermochemistry. Total energy analysis of the reaction pathway was performed by electronic structure calculations (ab initio and semiempirical methods) at different levels and basis sets [e.g., HF/6-311G(d), B3LYP/6-31G(d)]. Furthermore, the heterogeneous MnO2/PAA system was tested for the elimination of a model aromatic compound, phenol from aqueous solution. An artificial neural network (ANN) was designed to associate the removal efficiency of phenol with relevant process parameters such as concentrations of both the catalyst and PAA and the reaction time. Results were used to train and test ANN to identify an optimized network structure, which represented the correlations between the operational parameters and removal efficiency of phenol.