A relatively simple method of parameterizing subgrid-scale moist convection has been developed for use in tropical numerical weather prediction experiments. The technique includes entrainment, moist and dry downdrafts, and environmental mixing. Vertical extent of the model cloud is to one of three possible tropospheric levels. The formulation of the mass flux into cloud base is such that the amount of low-level energy consumed by the clouds is approximately that supplied by the large-scale flow. The model convection was employed, without change, in real-data primitive equation forecasts of two separate tropical disturbances. The first disturbance was a non-developing, Caribbean easterly wave (August 1961) and the second was hurricane Alma (August 1962). Predicted convection in the easterly wave tended to be weak and sporadic. Convective regions, generally in the eastern half of the wave, developed and then dissipated within 6–12 hr. Averaged over longer periods, latent heating tended to be weak and quasi-uniformly distributed over the eastern portion of the wave. The easterly wave showed no development in a 48-hr forecast. In comparison, predicted convection in hurricane Alma tended to be much stronger and more persistent. Most convective heating took place in the northeast quadrant of the disturbance, and the intensity of convection showed less temporal fluctuation. The strong and persistent heating led to intensification of the storm over the first 30 forecast hours and maintained a strong disturbance thereafter. Mean convective heating in Alma was approximately one order of magnitude larger than in the wave. Subsynoptic-scale bands and regions of convection predicted within the two disturbances agreed well with available radar and satellite observations. The predicted position of hurricane Alma was less than 90 statute miles in error at any one time during the first 36 hr of the forecast. The manner in which convection and boundary layer convergence occurred in the two forecasts represented some deviation from existing theories on conditional instability of the second kind (CISK). Two particular observations were made: 1) Ekman pumping was not a necessary or sufficient cause for the occurrence of cumulus convection; and 2) large-scale ascent not induced by cumulus latent heat release evolved as a significant factor in determining the total amount of predicted latent heating. Convection could persist for periods longer than 6–12 hr in a region only if the atmospheric destabilization by large-scale ascent was sufficient to counter the stabilizing effect of modeled convection.