The paper presents results of indexing methods evaluation of title and key words of textual documents in Serbian language. The following indexing method are evaluated.: — automatic indexing with single words, — automatic indexing with compressed single words and — manual indexing with key words. Evaluation is performed on the basis of the computation of a linear correlation coefficient between the actual relevance and the formal computer relevance. The actual relevance of two documents is computed by finding the set of common words in both of them. The formal relevance is obtained by means of finding of intersection of the document searching characteristic.