Speech production is a skilled motor task that is characterized by a remarkable variability. For a given sound, depending on the neighbor phonemes[1], on the prosodic structure [2], on the speaking rate or on clarity [3], the positions of the vocal tract articulators can be very different. This is due to a combination of factors, among which two are quite speech specific. First, the ultimate goal of speech production is not in a physical space, but in a cognitive one, the one of the linguistic information. The ultimate goal is categorical and allows variability within each category. Second, the relation between articulatory positions and perceptually relevant acoustic features is strongly many-to-one [5]. These degrees of freedom in excess enables an extraordinary capacity to compensate for many kinds of usual perturbations such as speaking with a pen between the teeth, with food in the mouth, by running.... This skill requires motor equivalence strategies. The present work is part of a larger project, that aims at studying how motor equivalence strategies are elaborated , and how they are used to deal with variations in the speaking conditions. Our methodology is based on the concept of uncontrolled manifold (UCM) [6]. The basic idea is that motor control variables can be divided into two major subsets: a subset in which variations directly alters task variables, and a subset, the UCM, in which variation has no (or very little) impact on the task variables. The UCM is mathematically defined as the null-space of the Jacobian of the relation between motor control and task variables. A biomechanical model of the tongue [7] was used, in which motor control variables are muscle length threshold as proposed by the Equilibrium-Point Hypothesis [8]. Ten French vowels have been studied. For each vowel the motor goal is specified as an ellipsoid in the space of the three first spectral maxima of the acoustic signal. For each vowel 20 sets of motor commands have been randomly selected. The variability of the UCM within each vowel category is assessed. Similarly the variability of the UCM across vowel categories is measured. These results are interpreted in terms of possible categorical representations of motor equivalence strategies across vowels, in particular in the light of classical phonetic features such as “front-back”, “open-closed” or “rounded-spread” vowels.[1] Daniloff, R. G., & Hammarberg, R. E. (1973). J. Phonetics, 1, 239–248[2] Cho, T. (2005). JASA, 117, 3867–3878[3] Matthies L.M., Perrier P., Perkell J.S. & Zandipour M. (2001). JSLHR, 44, 340-353[4] Atal, B.S., Chang, J.J., Mathews, M.V., & Tukey, J.W. (1978). JASA, 63, 1535-1555[6] Scholz, J.P., & Schöner, G. (1999). Exp. Brain Res., 126, 289-306[7] Perrier P., Payan Y., Zandipour M. & Perkell J. (2003). JASA, 114(3), 1582-1599[8] Feldman, A.G. (1986). JMB, 18, 17-54