An early established tenet of generative phonologists was that the SYSTEMATIC PHONEME should replace the neo Bloomfieldian AUTONOMOUS PHONEME in phonological theory, this being a gratuitous hypostatization frequently making it impossible to state phonological patterning in fully general form. This was early demonstrated by Robert Lees and Morris Halle in the now classical cases of Turkish and Russian devoicing. The present paper argues that for Tiberian Hebrew, ALTHOUGH autonomous phonemic analysis (2) and systematic phonemic analysis (3) clash in a considerably more radical way than do the above cases (4), NEVERTHELESS the distribution of the grapheme dagesh is interpretable only pursuant to the autonomous phonemic analysis (5). A solution is suggested to this apparent paradox (6).