It is a well known fact that the omission of referential subjects was licensed in Old French. Therefore, the grammar of Old French had to meet the relevant licensing conditions for null subjects. This licensing has been described for null subject languages such as Spanish and Italian in terms of an internal characterisation of the Agreement system which has to be either rich or strong depending on the definition of its content.

I will base my argumentation on Kato's (1999) proposal that a [+ pronominal] agreement system is the prerequisite for the licensing of null subjects. I follow her assumption that in null subject languages the agreement morphemes have the same grammatical status as a pronominal subject. As a result, they are able to check the EPP-feature of a given sentence via verb movement to AGRS. This account has two interesting consequences: First, it leads to an elimination ofpro, since it is assumed that the agreement morpheme itself takes over the functions which were attributed to the empty pronominal element; second, it makes the prediction that, for economy reasons, the projection of a specifier of AGRSP for EPP checking is excluded in null subject languages. However, any approach which is based on morphophonological criteria only fails to account for the licensing of null subjects in Old French in the 13th century, since their distribution is restricted to certain structural environments. Null subjects occur predominantly in main clauses with an initial non subject constituent and in conjunctional subordinate clauses which contain a preverbal topic. In order to capture this generalisation, one has to find out which additional