The observer-Kalman filter/eigensystem realisation algorithm is applied to the problem of generating a minimal state-space model realisation from a nonlinear helicopter model. The paper presents an interim result using a 10th order linear model, discusses the implications of this and generates some general guidelines before applying the technique to a nonlinear model. The outcome is a linear realisation which gives a better representation of the nonlinear model than obtained using small perturbation linearisation methods. The technique is capable of extracting unstable plant models from closed-loop stabilised systems, even when the feedback dynamics are not known.