This work deals with the stochastic inference of gas-phase chemical reaction rates in high temperature air flows from plasma wind tunnel experimental data. First, a Bayesian approach is developed to include not only measurements but also additional information related to how the experiment is performed. To cope with the resulting computationally demanding likelihood, we use the Morris screening method to find the reactions that influence the solution to the stochastic inverse problem from a mechanism comprising 21 different reactions for an air mixture with seven species: O2, N2, NO, NO+, O, N, e−. A set of six reactions, mainly involving nitrogen dissociation and exchange, are the ones identified to impact the solution the most. As such, they are assumed to be uncertain and estimated along with the boundary conditions of the experiment and the catalytic recombination parameters of the materials involved in the testing. The remaining 15 reactions are set to their nominal values. The posterior distribution is then propagated through the proposed boundary layer model to produce the posterior predictive distributions of the temperature and mass fraction profiles along the boundary layer stagnation line. It is identified that NO concentrations have the largest increase in uncertainty levels compared to cases where the inference problem is carried out for fixed chemical model parameter values. This allows us to inform a new experimental campaign targeting the reduction of uncertainties affecting the chemical models.