A procedure for initializing parameterizations for cumulus convection in numerical weather prediction models is described. The initialization adjusts the temperature and humidity fields so that a simplified version of the Kuo cumulus parameterization will yield diagnosed convective precipitation and vertical heating profiles if a specified velocity field can support them. In an unfavorable velocity field, the initialization will yield the closest approach to diagnosed convective precipitation possible. The initialization minimizes changes in the humidity and temperature fields while satisfying constraints imposed by the cumulus parameterization. Slight adjustments in the temperature field and relatively larger adjustments in the humidity field can modify the large scale from a state that does not support cumulus convection to a state whose convective heating, as parameterized by the simplified version of the Kuo scheme, agrees to the extent possible for an imposed velocity field. The use of more complicated versions of the Kuo cumulus parameterization with the intialized temperature and humidity profiles yields heating rates agreeing reasonably with diagnosed heating. If used in conjunction with an intialization for the velocity field, cumulus initialization may ameliorate problems associated with spinup of physical processes in numerical weather prediction.