In what sense can the sciences be said to constitute a (set of) specific cultural tradition(s) within broader culture? This is the proper way of posing the problem of the 'two cultures' today. For G. Bachelard the opposition between 'poem' and 'theorem' was fundamental, the elements of poetical imagination being radically different from the symbolical constructions of conceptual invention. In this article a sophisticated version of this point of view is proposed. The nowadays popular attempts to bridge all gaps between (...) the cultures by an appeal to creative imagination ('science as art', 'mathematics and poetry') are criticized. The real basis of considering research traditions nevertheless as cultural realities in a nontrivial sense is not to be found in a mediation between cognitive faculties, but has to be revealed by a comparative analysis of their respective products. In fact, science, as well as other cultural practices, is subject to variations of style (Granger, Hacking ...). Research styles are not in the first place due to external, sociological determinants, but to such factors as the necessity of individuating any idea in a work, the presence of changing images of science within each discipline, and the fundamental role of the not-yet-explicit, given the specific temporal structure of conceptual innovation with its „essential tension” (Kuhn) between the old and the new. (shrink)