The so-called passive in Māori has been the topic of a long-standing debate in the linguistics literature. Its frequency, especially in past tense narratives, makes this construction an atypical passive. It has been suggested that the passive in Māori is used with perfective (Clark 1973) and dynamic (Bauer 1997) events, and when the clause contains an affected direct object (Chung 1978). This thesis finds that all of these suggestions are correct, but, rather than a passive construction, it is ergative, so that Māori has split-ergativity. As predicted under the Transitivity Hypothesis (Hopper & Thompson 1980), the most transitive clauses in Māori have ergative marking, and less transitive clauses are accusatively marked. Transitivity is understood as a property of an entire clause, involving a number of factors, and the most important features of transitivity in Māori are PARTICIPANTS, AFFECTEDNESS OF O, ASPECT and PUNCTUALITY. Clauses that are low in transitivity are uniformly accusative, in both their morphology and syntax. However, highly transitive clauses, which we expect to follow ergative alignment, have some evidence of syntactic accusativity. This mixed behaviour follows directly from the Inverse Grammatical Relations Hypothesis (Manning 1996). Manning claims syntactic constructions like control, binding and imperative addressee are accusatively aligned in all languages, because they are restricted at argument structure. Languages can only be ergative at the level of grammatical relations, where syntactic processes such as relative clauses, question formation and topicalisation are restricted. It then follows that ergativity is only present in Māori at gr-structure in the most highly transitive clauses. We also look at Māori from a diachronic perspective, and see that it differs from its Eastern Polynesian sisters, which are all accusative. Māori is different because the extension of the imperfective pattern did not spread to all transitive clauses, thus preventing a reanalysis of imperfective clauses as active.