Abstract If there is any consensus in syntax at all, it is that syntax is all about tree structures —- so of course this is exactly what this talk will argue against on computational grounds. I will start with recent work in subregular syntax that actually supports the consensus: whereas the linear string patterns produced by syntax can be very complex (mildly context-sensitive), the tree patterns are much simpler and exhibit computational properties similar to what one finds in phonology and morphology (tier-based strictly local). But while syntactic structures may well be trees, syntactic constraints are different. I survey a variety of syntactic constraints, including island constraints, relativized minimality, freezing effects, and extraction morphology (e.g. wh-agreement). I show that these constraints form a natural class from a computational perspective, but only if they are computed over strings instead of trees. Crucially, these strings don’t encode linear order but specific types of prominence in the syntactic derivation. For example, the c[ommand]-string of a lexical item l contains all the lexical items that must have been selected later than l. The constraints above reduce to the same basic template applied to different types of strings. The overall system shares similarities with path equations in Lexical Functional Grammar yet is remarkably simple from a computational perspective (tier-based 1-definite). This greatly aids learnability and puts tight limits on the range of cross-linguistic variation we expect to see in those constraints. While syntax may well be about trees, it isn’t all about trees, and that’s actually a good thing.