Automatic labelling of continuous speech is an essential process in speech research to amount of time-aligned acoustic data. An automatic labelling method based on hierarchical representation of the energy function is presented in this paper. This method attemps to find a descending sequence of break points and does not proceed from the left to the right. First, the algorithm tries to label the consonants at the beginning of the sylables, secondly the vowels are labelled and finally the other phonemes are associated to the remaining signal The main idea of this method is to use the structure of the energy function to order the evolution of the labelling process.