Modern Hebrew has three kinds of genitive constructions. One of these, the Construct State (CS), sometimes—but not always—involves a unique phonological form of the head noun. An intuition shared by some speakers, stated explicitly by Siloni (2001, 2003), is that constructs where the head does not have a distinct CS form are sometimes less acceptable. Informal judgments, however, are extremely subtle and inconclusive in this case. The current paper uses corpus data to shed light on the question of whether such an effect actually exists. Using data from the heTenTen corpus and a dependency-parsed version of the Hebrew Wikipedia corpus, it is shown that in a large sample of genitives it is possible to observe an effect of overt CS marking. Specifically, the corpus data shows a higher proportion of CS in a sample of genitives having a higher proportion of morphologically variant head nouns. This is shown to hold both in a general sample of genitives, and in a sample restricted to deverbal nouns. Finally, we briefly explore the effect of head noun length on the choice of genitive, which illustrates another aspect of the role of weak phonological constraints on this alternation.