The frequency-domain nonlinear subspace identification (FNSI) method has recently been successfully applied to large-scale nonlinear structures. One of the key features of FNSI is the nonlinear generalisation of the stabilisation diagram. However, as in linear system identification, the selection of the model order in the diagram is complicated by the presence of spurious poles, resulting from noise and modelling errors. Spurious poles have been shown to strongly perturb the estimation of the no