Native language experience shapes how people perceive the world at an early age. Phonemic merger is a common phenomenon of language variation in which two phonemes cannot be distinguished in perception and pronunciation. Phonemic merger could indicate a deficit of perceptual sensitivity to non-native phonemic contrasts. For example, merged speakers from southwest China are always not able to distinguish the /n/-/l/ contrast that acoustically mixed in their dialects, whereas people from the north can distinguish them without an effort. A previous study (Conrey et al., 2005) showed that the perception of /I/-/E/ in English differ between merged speakers and unmerged speakers in the LPC response which attributed to the conscious phonological decision processes. However, it has been unclear that whether and how the language experiences of phonemic merger impact merged speakers’ phoneme perception in early auditory processing. Here, we adapted a passive oddball paradigm to examine if the merged experience impact speech recognition and explore the neural correlates of /n/-/l/ Chinese phonemic merger at the early stage of auditory processing, as indexed by the mismatch negativity (MMN). In the ERP experiment, the subjects sat in an acoustically shielded cabin. Acoustic stimuli /niú/ and /liú/ were delivered binaurally via headphones at an intensity of 60 dB SPL. We instructed the subjects ignore the auditory and concentrate on a silent movie. Simultaneously, we recorded the subjects’ brain waves. After EEG recording, all of the subjects participated in an /n/-/l/ discrimination test, in order to measure their capacity of distinguish /n/ from /l/. According to the accuracy of the behavioral test and assessment for dialectal experience, 22 listeners were divided into two groups, merged speakers (4 males, 7 females, age range, 18-25 years) and unmerged speakers (4 males, 7 females, age range, 18-22 years). Then, we compared the ERPs and behavioral performance for the discrimination of /n/-/l/ contrast between merged speakers and unmerged speakers. The results showed that the behavioral responses were slower (p = .003) and more error-prone (p < .001) in the merged speakers than in the unmerged speakers. These behavioral results indicate merged speakers’ deficit of discrimination of /n/-/l/ contrast. Our ERP results revealed that the deviant syllable /liú/ evoked a typical MMN only in unmerged speakers with a maximum amplitude peaked around 182 ms after the onset of the critical phoneme. The absence of MMN suggested that the /n/-/l/ merged speakers had lost the ability to distinguish the phonemes /n/ and /l/ in early auditory processing. Although failed to observe the MMN component in merged speakers on the MMN time window, we obtained a late discriminatory negativity (LDN) in 480~540 ms. For the precise function of LDN, more further studies are needed. The significant differences of ERP responses between two groups indicated that the merged speakers are deficient in recognizing the contrast between /n/-/l/ at least in the acoustic perceptual processing. In conclusion, our results indicate that long-term experience of phonemic merger affects merged speakers’ speech perception, and merged speakers lost the automatic discrimination of merged contrast at an early stage of auditory processing.