This paper presents a method for constructing and using transliteration rules which are based on Phonics, an instructional method for speaking and writing English letters. Conventional approaches to automatic transliteration often focused on statistical methods. However, the construction or the collection of correct transliteration examples is always the bottleneck of the statistical transliteration model. Also, in practical domains where the collection of such data is very difficult, such as education and tourism, it is reasonable to build a system without much qualified data. Furthermore, compared with Korean orthography of borrowed foreign words, the proposed approach is much easier to construct, and can generate more refined rules. The experimentation result shows that the proposed approach can improve the performance of a statistical-based transliteration system.