I provide some empirical evidence against models of adjectival modification involving functional sequences built on semantic classes (color, size, shape, etc.) and unrestricted (i.e. unprobed) movement. I argue that semantic constructs such as predicative/non predicative (or intersective/subsective, restrictive/non-restrictive, etc.) cannot be mapped uniformly (i.e. one-to-one) to specific adjectival structures. I argue in favour of using independently motivated syntactic structures to order adjectives (or modifiers). Specifically, I suggest that gradable adjectives are merged at the edge of the n phase and non-gradable adjectives in the complement of the n phase, yielding a gradable > non-gradable order. Some adjectives are freely ordered, because their complex internal structure renders them opaque to ordering principles.