In this article the description of an algorithm of a statement sentiment evaluation is done for the users' of social media language. We underline that statements of the natural language can be contradictory, emotionally complicated, ambiguous. It is the additional research task to detect the adequate formal criterion of the natural language statements ranging on t he scale 'posit ive - negative'. In the article the original decision of this problem on the base of the theory of the semantic field is described. The technique was tested in the All-Russian Scientific Research Institute of Labor of the Ministry of Labor and Social Protection of the Russian Federation for investigation of population opinion to the new forms of employment. Empirical base is more than 100 000 messages of users of thematical groups in VKontakte. Analysis with the accent on the parameters: subject of tonality, object of tonality, message tonality was done. The technique of research assumes the detection of words-markers that indicate the general message tonality. © 2019 FRUCT.