Abstract Holmberg (Holmberg, Anders. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4). 533–564) and its revised version in Holmberg et al. (Holmberg, Anders, Aarti Nayudu & Michelle Sheehan. 2009. Three partial null-subject languages: A comparison of Brazilian Portuguese, Finnish, and Marathi. Studia Linguistica 63(1). 59–97) derive the availability of null subjects in a given language from the interaction between T with/without a D(efiniteness)-feature and the features of subject pronouns. Their theory predicts the existence of consistent null subject languages, whose T has the D-feature, and partial null subject languages, whose T lacks the D-feature. This paper examines this D-feature approach to null subjects against the empirical evidence provided by Brazilian Portuguese, a partial null subject language, and European Portuguese, a consistent null subject language, showing that it cannot account for the range of microvariation observed with respect to different null subject pronouns and the type of T (finite vs. participle vs. gerund). We argue that, in comparison, the ellipsis account of null subject licensing put forward in Martins and Nunes (Martins, Ana Maria & Jairo Nunes. 2021. Brazilian and European Portuguese and Holmberg’s 2005 typology of null subject languages. In Sergio Baauw, Frank Drijkoningen & Luisa Meroni (eds.), Romance languages and linguistic theory 2018. Selected papers from “Going Romance” 32, Utrecht, 171–190. Amsterdam & Philadelphia: John Benjamins) fares better. It retains from Holmberg (Holmberg, Anders. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4). 533–564 et seq.) the insight that the licensing of null subjects depends on the interaction between the features of T and the features of subject pronouns but resorts only to ϕ-features and Case. Crucially, it relies on the (theoretically and empirically) plausible assumption that the relation between abstract ϕ-features and verbal agreement morphology need not be transparent.