This paper deals with the subspace identification of unstable transfer function models with numerator dynamics and time delays. The unstable transfer function models are simulated and identified by the present method for both deterministic and stochastic cases. Two case studies are considered for evaluating the performance of the present method. The quality of the identified model is tested by mean relative square error and mean-related variance indices. The problem associated with the identification of unstable systems corresponding to stable systems in terms of sampling time, need for improved filter, achievable performance evaluation and validation is brought out. The simulation results are compared by means of Bode plot and estimated poles with the real pole of the true plant. The model is validated through the residual analysis. A computer programme (in MATLAB) is also provided. © 2015 Indian Institute of Chemical Engineers.