Abstract Liaison is a sandhi phenomenon in French. Over the last four decades, it has given rise to many different models illustrating the whole range of phonological theories. More recently, new studies have documented its acquisition in French-speaking children as well as adult learners of French as a second language. These studies have resulted in the elaboration of two models of the acquisition process: (1) the constructionist model (Chevrot et al., 2009; Nicoladis and Paradis, 2011) developed within the framework of the usage-based theories; (2) the phonological model (Wauquier, 2009) which represents the framework of nonlinear phonology. Our aim is to re-examine the usage-based model in the light of the criticisms and suggestions made by Wauquier (2009). We shall first present the two models and then examine the issues under discussion. After that, we shall present longitudinal data testing a prediction made by the phonological model with regard to the generalization process in L1 and L2 acquisition. To conclude, we shall identify the points that remain to be clarified for each of the models and the directions which future research should take.