This study investigates how seven Chinese sentence-final particles (SFP le, ne1, ma, ne2, ba1, ba2 and a; hereafter SFP) and their features are represented in English speakers’ L2 Chinese. In this research, SFPs are analysed as heads instantiating different positions in the CP domain (Paul 2009, 2014, 2015), which are head-final, and in particular, they are considered to carry semantic, syntactic and discourse features. As there is no SFP in English, the features on Chinese SFPs are realised by a variety of syntactic means. Through a proficiency test and six experimental tasks, data from 76 participants (including 18 Chinese native speakers, 20 low-intermediate learners, 20 high-intermediate learners and 18 advanced learners) were collected. Results show that English-speaking L2 learners can easily establish the basic syntactic structure of Chinese SFPs and successfully acquire the features attached to SFPs ma, ba1 and a. However, they have significant difficulty in acquiring the features attached to SFPs le, ne1, ne2 and ba2. In general, syntactic features on Chinese SFPs are intact in L2 grammars, whereas semantic features (i.e. syntax-semantics interfaces) are very vulnerable. In addition, it is found that not all discourse features (syntax-discourse interfaces) are problematic. Findings indicate that both L1 grammar (i.e. L1 transfer) and L2 input (frequency, saliency and complexity) play important roles in affecting learners’ acquisition of the features attached to Chinese SFPs. In particular, learners seem to transfer all feature sets from their L1 English. Semantic features that have been transferred from their L1 English but that are neither confirmed nor disconfirmed by the Chinese input have become dormant in the L2 Chinese, which complements the Dormant Feature Hypothesis (Yuan 2014). Furthermore, the homomorphous SFPs which exhibit a ‘one-to-many’ form-meaning connection are believed to complicate learners’ recognition and acquisition of relevant features on SFPs. It is also demonstrated that the mapping of a feature across CP domains (i.e. from a two-CP structure to a one-CP structure) can be problematic and difficult. The discourse feature needs to be reassembled in L2 grammars, which advances the arguments of the Feature Reassembly Hypothesis (Lardiere 2008, 2009a,b).