A rule based graph-theoretical system has been used to evaluate qualitatively the activity of a class of nucleoside analogues against human immunodeficiency virus (HIV). The system identifies biologically relevant vertices (atoms) in the molecular graphs of the compounds which have the biological activity of interest. The idea is to relate biological activity with the structural or substructural characteristics of the compounds from the point of view of molecular topology (connectivity). The system brings vertices of similar or close topological environment in the respective compounds together and this is reflected in the ranges of values formed by a distance based index of the vertices, the 'distance exponent index (Dx)', where x is any real number. It is found that the system makes correct prediction of the activity of all the compounds (active as well as inactives) of both training set and the test set against HIV. It is also apparent from this study that the index D-4, which has been used here, can make a useful classification of the vertices according to their molecular environment and the system can produce significant result in a small as well as diverse data base.