This study aims to test possible transfer sources at the initial stage of third language (L3) acquisition by examining the acquisition of four Mandarin sentence-final particles (SFPs) by Cantonese-English bilinguals. Unlike the case in English, in both Mandarin and Cantonese, a statement can be converted into a question by merging an SFP at a sentence final position. Cantonese is therefore structurally closer to Mandarin than English and is thus predicted to be the source of transfer on the basis of the Linguistic Proximity Model (LPM; Westergaard et al., 2016) and the Typological Primacy Model (TPM; Rothman, 2010, 2011, 2015). An offline Acceptability Judgement Task (AJT) and an online Cross-Modal Priming Task (CMPT) were employed. The AJT results show that the learners’ Cantonese facilitated them in acquiring the features of the Mandarin question particles, which supports the predictions of the TPM and the LPM. In addition, the online processing data indicate that the L3 learners’ implicit knowledge developed later than their explicit knowledge.