This paper focuses on NumP, an independent projection for number,which was proposed in pre-minimalist research (Picallo, 1991; Ritter,1991). Arguments for the relevance of maintaining it as anindependent projection in a minimalist framework, contrary toChomsky (1999), who assumes number is one of the j-featuresassociated to N, are provided. The consequences such presenceimposes for DP internal agreement phenomenon are discussed andsome alternatives to account for it are offered.