Of all the correspondance-based faithfulness constraints in Optimality Theory (McCarthy and Prince 1995), perhaps the least scrutinized has been the constraint which regulates metathesis: Linearity. This paper presents formal and empirical reasons in favor of replacing this constraint with other faithfulness constraints. Formally, it is shown that Linearity is categorically more powerful than most other constraints phonologists employ in their analyses. The empirical evidence is drawn from Kwara'ae (Austronesian), whose unusually robust process of CV metathesis highlights the inability of a single faithfulness constraint regulating linear order to account for three properties of CV metathesis. The proposed solution replaces Linearity with a family of constraints based on the notion of contiguity. These contiguity constraints are sufficient to account for the language data, but lack the formal power associated with Linearity.