An approach based on the Nonlinear Autoregressive Moving Average Model With Exogenous Inputs (NARMAX) is used to analyze simultaneous measurements of the geomagnetic Dst index and VBs, the merging rate of the interplanetary magnetic field and the geomagnetic field. A nonlinear model which describes the dynamics of the Dst index driven by VBs is identified directly from experimental measurements. The identified model is then used to forecast the evolution of the Dst index and to compute generalized frequency response functions (GFRFs). Models were also identified on different intervals of data to check the stability of the main properties of the GFRFs. Analysis of these functions provides information about the types of nonlinearities involved in the energy storage process in the magnetosphere; in particular, the importance of coupling processes between various spectral components in the input which transfer energy to the summation frequency in the Dst index is illustrated. It is shown that the GFRFs derived from the data can be used to validate theoretical models which have been proposed to describe the evolution of the magnetosphere. For example, a coupling between various spectral components in the input indicates that linear models which relate VBs to Dst are inadequate to represent real physical processes involved in the Dst evolution. ¿ 2001 American Geophysical Union