The problem of rainfall prediction for a tropical storm over the United States is examined from a diagnostic point of view. A moisture budget is constructed and the differences between the computed and observed precipitation are discussed. Although the areal averages were comparable, the point-by-point agreement was only fair. Stable and convective precipitation are then computed using methods common to many numerical models. The discrepancies between forecast and observed precipitation is assumed to be due to the incomplete formulation of the amount of moisture available for convection. This leads to an expression for the subgrid-scale moisture supply that provides the missing precipitation. Methods are then suggested in which the deficiencies of a parameterization scheme could be corrected during its use in a prognostic model.