This work focuses on electromyographic (EMG) signal processing. We propose a new blind approach that aims at detecting the firing rates of the activated motor units. The proposed method is based on the fact that, EMGs can be modelled as second-order cyclostationary signals. After application of a Blind Source Separation (BSS) algorithm, we compute a cyclostationarity measure which is the Cyclic Spectral Density (CSD), and we show how one can use it to group the estimated components into independent subspaces and in an automated manner. The proposed classification procedure is based on the concept of subspace BSS techniques, like the multidimensional Independent Component Analysis (MICA), the difference being that our method allows automatic classification of the estimated source signals. After discarding the subspace corresponding to the noise and computation of a modified CSD measure, the proposed procedure yields to the detection of specific cyclic frequencies corresponding to the discharge frequencies of the Motor Units Action Potential Trains (MUAPTs). Early results obtained from experiments on synthetic EMGs are presented in the paper and research perspectives conclude this work.