Abstract: In order to identify language rapidly and effectively, a method for CV-syllable extraction based on the Vowel Onset Points (VOP) detection is proposed, on this basis, a new method for language identification based on CV-syllables is researched. First, a double-threshold endpoints detection from both sides is given to get speech segment, which could avoid great risk of wrong decision. Second, VOP are detected by Linear Prediction Residue Error (LPRE) and the CV-syllables from the speech segment are obtained exactly. At last, feature vectors for each CV-syllable are extracted. The Support Vector Machine (SVM) is adopted to realize language identification. The simulation experiment for English, Mandarin and Cantonese shows that VOP detection makes sure the precision of each CV-syllable extraction. The new method has the high correct response rate. The change of CV-syllable length almost has no effect on identification results. The training time for model is so short that language identification could complete efficiently.