A collection of scholarly journals must adapt to the needs of research through annual suppressions and additions. Several methods have been proposed : interviewing researchers, usage studies and interlibrary loan statistics, which are difficult to put to work and whose results are partial and not always satisfactory. More recently, the use of the « impact factor » indicating the frequency of citation does not produce results as clear as its advocates affirm. This proves the ambiguity inherent in the notion of the « impact factor ». The study of citations themselves, outside of the impact factor, seems to give a better basis for the work of adapting collections.