In this paper we deal with large sample test for coefficient of variation when the observations come from a normal population. Such large sample tests are based on normal approximation for sample coefficient of variation. When the sa mple size is small or moderate, the normal approximation may not be accurate in the sense that the attained level of the test Is not close to the nominal level. Following Bhattacharya and Ghosh (1978), we obtain valid Edgeworth expansion for sample coefficient of variation to O(n-¹) under simple null hypo thesis and contiguous alternative hypothesis. This helps one to determine the critical region such that attained level of significance is closer to the normal level and the power function is more accurate.