Abstract. The isomorphism invariance criterion of logical nature has much to commend it. It can be philosophically motivated by the thought that logic is distinctively general or topic neutral. It is capable of precise set-theoretic formulation. And it delivers an extension of 'logical constant' which respects the intuitively clear cases. Despite its attractions, the criterion has recently come under attack. Critics such as Feferman, MacFarlane and Bonnay argue that the criterion overgenerates by incorrectly judging mathematical notions as logical. We consider five possible precisifications of the overgeneration argument and find them all unconvincing.The standard approach to logical consequence in the modern literature is the post-Tarskian model-theoretic conception. On this conception, a sentence φ is a logical consequence of some set of sentences Γ just if every model of Γ is a model of φ. A model consists of a nonempty domain of objects and a valuation function that assigns semantic values to the nonlogical expressions of the language. As such, the standard approach relies on a distinction between logical and nonlogical expressions. It is striking, then, that there is little agreement over how this crucial distinction should be drawn. We defend the most natural supplement to the model-theoretic definition: isomorphism invariance.The isomorphism invariance criterion of logical nature has much to commend it. It can be philosophically motivated by the thoughts that logic is distinctively general or topic-neutral. It is capable of precise set-theoretic formulation. And it delivers an extension of 'logical constant' which respects the intuitively clear cases, namely identity, the Boolean connectives and the universal and existential quantifiers-the logical constants of first-order logic with identity.Despite its attractions, isomorphism invariance faces a major recurring objection in the literature.