The areas affected by the L1 transfer to L2, such as phonotactics and stress, appear to have different degrees of erasability through training and practice. This paper attempts a theoretical analysis to account for empirical observations of L1 transfer in diphthong-coda coalescence and in sentence stress for Chinese learners of English in an American accent training course. Using the framework and tableau method of Optimality Theory（Prince and Smolensky 1993）, the phonological paradigm that views languages’ differences as symptomatic of differences in rankings of a set of universal violable constraints, this paper posits the underlying ranking at work in（a） the segmental case where [a?] is the observed output for/a?n/;（b） the suprasegmental case where no stress contrast is observed for a pair of phonologically similar sentences with contrastive stress. It is argued that, in the segmental case, the coalescence of the second vowel of a diphthong with an alveolar nasal coda results from the high ranking of markedness constraints, including Complex Syllable; and in the suprasegmental case, sentence stress is greatly affected by changes in the ranking of the constraint culminativity. Aside from shedding light on the understudied issue of diphthong-coda coalescence, this research also contributes to the discussion of the comparative erasability of L1 transfer for the two areas and demonstrates how a deduction-based theoretical method may actually have practical implications for guiding the design of L2 teaching methodology of English pronunciation as well as for promoting native English speakers’ clearer understanding of Chinese learners of English on the global stage.