Abstract : This mainly experimental study aims at determining the most relevant acoustic and prosodic parameters of emphasis in English spontaneous speech. In order to decide which words or passages are emphatic in the data base chosen, a perception test based on naïve native speakers is carried out, which makes it possible to determine a degree of emphasis for each word. The data base is labelled into rhythmic units (intonative units, tonal units and words) and phonetics. An acoustical analysis is then performed, based on the hypothesis that fundamental frequency, segmental duration, global intensity and pauses are certainly essential in the perception of emphasis. The hypotheses are verified and parameters which seem relevant for a statistical analysis are chosen thanks to the acoustical analysis. A program for tree-structured analysis is used on the segments of the data base and also on the data for each speaker, and for each type of discourse in order to see if different strategies can be extracted. Finally, a new experiment based on manipulated synthethised speech parallels the statistical analysis. Fundamental frequency, segmental duration and pauses are modified in turn in certain segments of the data base. These segments are also delexicalised and a new percepion test is carried out, in which the semantic criterion is integrated, which is not the case in the statistical analysis. As a conclusion, emphasis is a way to highlight something showing more-than-normal involvement by increasing the energy spent to utter the emphatic occurrence, and/or increasing the pitch and/or the melodic movement on this occurrence.