Abstract: Nonlinear predictive error of a time series can be used to distinguish chaos from randomness for a dynamical system which generates the series. Based on this, we analysed a time series of AEindex (from 12h, Jan. 30, 1982) with a resolution of one min and 15000 data points. At first, using one part of the data, a wide predictive model is constructed. The model varies from nonlillear deterministic extreme to the linear stochastic extreme. Then another part of the data is Predicted by this model. The results of predictive error show that the magnetospheric process discribed by AEindex displays a chaotic behaviour. Using the same data and the method of time delay reconstructing phase space, the correction dimension of the attractor is computed, it is about 2.5.