This paper presents evidence that constraints are expressed abstractly, in terms of formal phonological categories, rather than directly encoding phonetic information.  The argument is made on the basis of constraints that have the same functional basis, but distinct formal properties: ONSET and *ONSET/X.  If phonetic factors are projected directly onto phonological constraints, then the shared functional basis of these constraints should entail that they cannot be formally distinct.  Specifically, the direct-phonetics model predicts that ONSET is actually '*ONSET/zero,' a constraint against null onsets that is the highest ranked member of the *ONSET/X family.  However, there is phonological evidence against equating ONSET with *ONSET/zero.  First, ONSET can be freely ranked with respect to *ONSET/X constraints, although the *ONSET/X constraints are themselves in a universally fixed relationship derived from the sonority scale.  Second, the behavior of glide-initial syllables in languages that avoid high-sonority onsets while also banning onsetless syllables shows that ONSET and *ONSET/X evaluate different phonological structures.  Each of these factors independently demonstrates that ONSET and *ONSET/X are formally distinct.  Consequently, even phonetically grounded constraints like these do not encode phonetic information directly.