The legal restriction of hate speech – i.e. speech that expresses contempt for people on the basis of their ethnicity, religion, or sexuality – is now commonplace in liberal legal systems outside the United States. This thesis takes up the question of whether restrictions on hate speech are generally justifiable. I begin by explaining why liberals should not dismiss anti-hate speech law from the outset as an intolerable violation of free speech. My analysis of the case for anti-hate speech law is thereafter framed by two main concerns. Firstly, I stress that if we are to impose legal restrictions on hate speech, we must establish not just that there are harmful outcomes associated with hate speech, but that those who engage in hate speech are responsible for those outcomes. Secondly, I argue that restrictions on hate speech should be assessed in two distinct classes. Inquiries into the justificatory bases of anti-hate speech law are typically conducted as if informative generalisations can be made about how the law should respond to anything that is properly called hate speech. Against this approach, I argue that while the liberal state can and should impose restrictions on directly harmful hate speech (in which hate speech is used to threaten, harass, and incite violence), restrictions on indirectly harmful hate speech – in which hate speech (allegedly) contributes to identity-based social hierarchies and their concomitant harms – are not justifiable. The problem with restrictions on indirectly harmful hate speech is not the structure of the liability-ascription framework under which they operate. Rather, I argue, the problem is epistemic: we cannot confidently judge that hate-speakers are in fact responsible for contributing, more than trivially, to the harmful patterns of identity-based inequality and disadvantage in light of which restrictions on indirectly harmful hate speech may be defended in principle.