The paper discusses a problem concerning theoretical validity of sociological measurement. The author analyzes historical development of the validity category mentioning a scientific approach to the problem taken by sociologists of the former Soviet Union. Special attention is paid to the notions of 'concept' and 'construct' as related categories. The author suggest comparing these notions in three ways: (1) based on the degree of abstractedness; (2) referring to the degree of verificability; (3) with regard to various interpretations given to both notions. The paper explores validity as a concept, and as a construct. Taking into account W.Trochim's model of conceptualization, the author introduces his own approach to the validity of sociological measurement. There are five types of validity obtained in sociological research: theoretical, construct, internal, external, and predictive. Besides, the article also provides a detailed analysis of the structure of theoretical validity and its kinds (face, content, composition, and logic validity). Face validity con¬sists in appropriateness of theoretical construct to a social phenomenon brought into focus of investigation. The essence of content validity is to give a thorough description of theoretical construct, while composition validity deals with structure of theoretical construct and its dynamics. Logic validity refers to further operationalization of the phenomenon being studied. The main points discussed in the paper are data biases. The author explains what kinds of biases may occur by theoretical validation procedure and gives concrete examples of possible consequences. In conclusion, specific features of the relationships between all types of validity are schematically presented.