This study examines relative weighting of two acoustic cues, vowel duration and spectra, in the perception of high front vowels by Japanese learners of English. Studies found that Japanese speakers rely heavily on duration to distinguish /i?/ and [character omitted] in American English (AmE) as influenced by phonemic length in Japanese /ii/ and /i/, while spectral cues are more important for native AmE speakers. However, little is known as to whether and how this non-native perceptual weighting can change as a result of L2 learning. By employing computational and experimental methods, the present study shows that Japanese learners of English exhibit different cue weighting depending on which language they think they hear. The experiment shows that listeners use more spectral cues and less durational cues when they think they are listening to 'English' stimuli as opposed to 'Japanese' stimuli, despite the stimuli being identical. This result is generally in line with our computer simulation, which predicts distinct developmental paths in first language (L1) and second language (L2) perception. The Second Language Linguistic Perception (L2LP) model, which incorporates the language mode hypothesis, provides a comprehensive explanation for the current findings.