AbstractWe examine the middle atmospheric odd‐hydrogen (HOx) chemistry by comparing the Aura Microwave Limb Sounder (MLS) OH and HO2 measurements with a photochemical model simulation. The model underestimates mesospheric OH and HO2 concentrations if the standard chemical kinetic rates are used, whether the model H2O and O3 are constrained with observations or not. To resolve the discrepancies, we adjust the kinetic rate coefficients of three key reactions (O + OH → O2 + H, OH + HO2 → H2O + O2, and H + O2 + M → HO2 + M) and the O2 photo absorption cross section at Lyman α (121.57 nm) using the Bayesian optimal estimation. A much better model‐observation agreement can be achieved if the kinetic rate coefficients for H + O2 + M → HO2 + M is increased by 134–310%, and the O2 photo absorption cross section at Lyman α is reduced by 33–54%, while the kinetic rate coefficients for O + OH → O2 + H and OH + HO2 → H2O + O2 remain consistent with the current laboratory values. The kinetic rate coefficient for H + O2 + M → HO2 + M requires a very large adjustment beyond the uncertainty limits recommended in the NASA Data Evaluation, suggesting the need for future laboratory measurements. An alternative explanation is that the radiative association reaction, H + O2 → HO2 + hν, plays a significant role, which has never been measured. Our results demonstrate that high‐quality satellite observations can be used to constrain photochemical parameters and help improve our understanding of atmospheric chemistry.