Matrix factorization algorithms are increasingly used to extract meaningful information from multivariate EMG datasets. However a key issue is the selection of the number of synergies (i.e., model order) to retain. In this preliminary work a set of criteria, based on Independent Component Analysis, was developed to determine the number of synergies to extract from a multivariate EMG dataset, and applied on EMG signals acquired from 12 leg muscles during walking at different cadences (40, 60, …, 140 strides per minute) in young and elderly subjects. The method was tested on ad-hoc created datasets with a predetermined number of embedded sources and amplitude of added noise. Young subjects walking patterns are explained by a number of synergies not significantly different with respect to elderly subjects. The inter-subject variability is greater at high (elderly) and low (young and elderly) cadences suggesting that the walking pattern is more stable at central frequencies. The type of preprocessing influences the number of underlying synergies: an increased number of independent components is needed to explain the variability of unfiltered data. The proposed method could serve as a guideline to scientists in the evaluation of walking performance. Further developments will include a validation of the method and its extension to other factorization algorithms.