Our goal is to recover articulatory information from the speech signal by acoustic-to-articulatory inversion. Like most inversion methods proposed in the literature, our method relies on the analysis-by-synthesis paradigm, here based on Maeda's articulatory model. After an overall description of the inversion method the paper presents how inversion can be used to investigate acoustic properties of the articulatory model, which helps us to formulate the way of incorporating effective constraints to obtain phonetically realistic inverse solutions. First, we improved the articulatory codebook construction by achieving a better approximation of the articulatory space boundaries. Then, we adapted geometrical parameters of the articulatory model to guaranty a better acoustical faithfulness with respect to the original acoustical data measured. Finally, we show some inversion results obtained on vowels uttered by the female subject whose images were used to build the articulatory model.