ER-2 aircraft observations of OH and HO2 concentrations in the upper troposphere during the NASA/STRAT campaign are interpreted using a photochemical model constrained by local observations of O3, H2O, NO, CO, hydrocarbons, albedo and overhead ozone column. We find that the reaction O(1D)+H2O is minor compared to acetone photolysis as a primary source of HOx(=OH+peroxy radicals) in the upper troposphere. Calculations using a diel steady state model agree with observed HOx concentrations in the lower stratosphere and, for some flights, in the upper troposphere. However, for other flights in the upper troposphere, the steady state model underestimates observations by a factor of 2 or more. These model underestimates are found to be related to a recent (<1 week) convective origin of the air. By conducting time-dependent model calculations along air trajectories determined for the STRAT flights, we show that convective injection of CH3OOH and H2O2 from the boundary layer to the upper troposphere could resolve the discrepancy. These injections of HOx reservoirs cause large HOx increases in the tropical upper troposphere for over a week downwind of the convective activity. We propose that this mechanism provides a major source of HOx in the upper troposphere. Simultaneous measurements of peroxides, formaldehyde and acetone along with OH and HO2 are needed to test our hypothesis. ¿ 1997 American Geophysical Union