A new method of evaluation of similarity between two nucleotide sequen
ces is proposed. It is based on comparing frequency/correlation dictio
naries of the sequences under investigation. The dictionary is a set o
f all strings of various length occurring within the sequences accompa
nied by their frequencies. The method proposed allows to compare seque
nces of arbitrary lengths, its advantage is absence of necessity of in
formal (expert) choice of the best fit. Efficiency of the method is de
monstrated by comparison of several human genes and viruses.