AbstractA nonlinear systems analysis is applied to the tracks of 249 tropical cyclones (with a six‐hour sampling time) in the Australian tropics for the period 1959–1980. First estimates are obtained of the degree of their chaotic, or irregular, behaviour. The degree of chaos is estimated by normalizing all trajectories to a common initial position and measuring the average rate at which initially close pairs of pieces of trajectories diverge.It was found from the correlation integrals calculated for the tropical cyclone tracks that the dimensionality of the underlying dynamical processes appears to be between six and eight, and that the time scale for e‐folding error growth is about one day. The results of this study therefore suggest that the movement of Australian region tropical cyclones is predictable deterministically up to about 24 hours in advance. Beyond that limit, consideration must be given to statistically based techniques.These findings were supported further by comparing the rate of growth of the observed Australian region tropical cyclone position variance with that derived from a random walk model superimposed on a mean drift. The correspondence was very close, with both the empirical and theoretical position variances growing linearly in time after approximately the first 18 to 24 hours, confirming that stochastic models have a role to play in forecasts beyond 24