This paper presents a critique of Sagart's (2004) classification of the Formosan languages. Sagart proposes a subgrouping based on a set of innovations in the numeral systems of the Formosan languages. These innovations entail that the higher phylogeny of Austronesian is much more hierarchical than in other subgrouping accounts (e.g., Blust 1999, Ross 2009). According to Sagart, the innovated numeral forms can be derived from the complex numerals of the Formosan language Pazeh. A number of arguments are presented that call this subgrouping account into question. There are reasons to assume that the proposed numeral derivations are highly unlikely, and that similarities between the complex numerals of Pazeh and the innovated forms in other Formosan languages are due to chance. Furthermore, language contact might account for the pattern of numeral innovation. Finally, the migration pattern that is entailed by Sagart's numeral innovations does not take into account possible back and forth migrations of the Formosan tribes.