The purpose of this study is to investigate the developmental path of high vowel devoicing (HVD) in Japanese. A picture-naming task was conducted with Japanese-learning preschoolers of three and four years old. The empirical data presented in this study allow us not only to make comparisons with the data from 4 year-olds in a previous study, but also to address the devoicing patterns for the 3-year old children, which have received little attention in the developmental literature on HVD. The results of twenty children reveal distinct patterns depending on position; Word-medially, the overall occurrence of HVD increases if we compare the average HVD rates between the ages of 3 and 4, but their rate is not yet reached at the adult-like level at the age of four. Word-finally, on the other hand, the rates are overall lower than the word-medial devoicing. Further, unlike the incremental pattern observed in the rates for the word-medial devoicing, no clear developmental advancement is found in this position. The presence or absence of the developmental advancement appears to support the qualitative differences between two types of HVDs in distinct positions.