This article looks at two current models of feature geometry, the Halle-Sagey model as modified by Halle (2005) and the Clements-Hume model, from the perspective of palatalization and related processes in Polish, Russian, and Ukrainian. The Halle-Sagey model predicts that palatalization should be analyzed by assuming derivational levels and is thus at odds with the tenet of strict parallelism in Optimality Theory. In contrast, the Clements-Hume model appears to be able to achieve the same goal without recourse to derivational stages because it is based on the assumption that, in the ways relevant for palatalization, vowels and consonants are characterized by the same features. However, analysis of palatalization and related processes shows that this assumption is incorrect. The consequence is that derivational stages cannot be avoided and that the tenet of strict parallelism must be rejected.