Abstract : Do bright stars have secular brightness variations ? Although this problem has been studied by some astronomers during the last centuries, no serious answer has been given to it. A hundred years ago, C. Flammarion collected old stars brightness estimates and published them in his book Les étoiles et les curiosités du ciel. It was obvious to him that a great number of stars had secular variations. Pickering (1895) and Zinner (1926) published two other compilations of catalogues. They did not agree with Flammarion's opinion, but they were not really able to analyze the amount of genuine information the old data contained. No other study has been published since then. New data analysis methods have been created in the last years, such as Correspondence Factorial Analysis, that allow further studies of large data tables. The aim of the work here presented is to appraise whether those methods can help to progress in this secular variations problem. The use of original sources allows to build a table of data integrating "signatures" of suspected causes for the magnitude différences (from astrophysical causes to data reliability problems). Factorial analyses of this table have be performed, thanks to new data coding methods, as "differences with expected values" coding or "unsharpened" coding, created to get rid of notation differences between observers. It is not possible to find significatively correlated "signatures" and expected effects, such as color observers' equations, are not visible. It seems that there is a kind of global brightness "memory", from one catalogue to an other, except for Sir William Herschel's catalogue which is also the only one to give brightness comparison and not magnitudes values. Perhaps has it been protected from this "memory" effect by this notation difference ? This work does not pretend to close the debate about secular variations. It ends with the description of the further inquiries that are to be done : in order to draw astrophysical conclusions, it can be then useful to test some statistical treatment combining factorial and Fourier analyses. Such treatment could allow to get rid of phase information and open the way for periods correlation studies.