This paper proposes a new objective methodology for analysing the verbatim transcripts of central
bank announcements. I apply a statistical method based on locally recurring word patterns, the
descending hierarchical classification algorithm, to identify the characteristic themes of the monthly
press conferences of the President of the ECB. The econometric evidence suggests that these press
conferences convey information to the public. Moreover, the reaction of market rates strongly depends
on the principal themes of the statement.