Abstarct: Human auditory system can recognized the same vowel pronounced by different people with different gender, different age. However this is a difficult task for machine to do it accurate. Many other transforms with different properties have been devised in order to make certain operations easier or certain features more easily visible. Among these, the Mellin Transform, and its restricted version called the Scale Transform, can represent a signal in terms of scale. The key property of the scale transform is the scale invariance. This thesis proposed a new feature extraction algorithm for vowel modeling and recognition using scale transform and one of the known methods for feature extraction such as Mel frequency Cepstral Coefficient (MFCC). Experimental results indicate that the new features produced better accuracy than original feature extracted from the Mellin transform and other methods in application of vowel recognition.