Markov switching models have received increasing attention in time series analysis due to their ability to detect latent states in the dynamics of the analyzed variables. Generally, the number of states is fixed a priori because it is impossible to apply classical statistical tests, due to the problem of disturbance parameters present only under the alternative hypothesis. In the present contribution we show, via Monte Carlo simulations, that fuzzy clustering can mimic the parametric inference on the regimes derived from the estimated model; furthermore, typical indices used in clustering to determine the number of groups can be used to identify the number of states. The procedure is very simple, considering that it is performed (non-parametrically) independently of the data generation process and that the validation criteria we use are available in most statistical packages. A final application on real data completes the analysis.