That demonstrative thoughts depend on perception for their content seems palatable to many philosophers. Allegedly, the thoughts preserve, as content, the objects and properties that perceptual discrimination picks out. I shall argue that for that to happen, nonconceptual modes of presentation would have to be an essential part of demonstrative thoughts: they would have to enter the truth-conditional content of those thoughts. Otherwise, the content of perception won't enter the thoughts. But this is a far less palatable idea than the one we started out with: how can anything be a thought unless it's composed entirely of concepts? I'll defend the view that nonconceptual perceptual modes of presentation can play roles as constituents of certain thoughts. Specifically, they allow subjects to determine extensions and do not necessarily violate generality and compositionality constraints on thought. Nonetheless they are not conceptual representations, because they are entirely stimulus-dependent and cannot be memorized to support de-identification.