Developing the adjoint model by hand is tedious, time-consuming, and error-prone work, especially for a large model such as ARPS, which is composed of more than two hundred subroutines. We therefore use ADIFOR to compute sensitivities of all given dependent variables (DVs, e.g., forecast aspects and their diagnostic functions) with respect to all given independent variables (IVs, e.g., initial and boundary conditions). For a single run, ADIFOR performs one control run and as many TLM runs as the number of IVs, implicitly. The final results are exactly the same as would be obtained from as many ADJM runs as the number of DVs. In meteorology, ADIFOR has been applied to a 1-D convective cloud model [17, 16], a 3-D storm-scale model [15], a mesoscale model [5], and an air quality model [11].