It has become clear over the past few years that the standard LFG treatment of agreement is inadequate, a fact which was the focus of the agreement workshop at the LFG 05 conference. This inadequacy is a combination of several factors, but it is primarily a conceptual failure. An adequate theory of agreement must begin with a conceptual understanding of the nature and purpose of agreement. This paper will argue that agreement, along with Case marking, involves an additional level of representation in the LFG projection architecture, g-structure (grammatical marking structure). G-structure is conceived of as a projection from f-structure, but its separate identity from f-structure allows it to more faithfully express the kinds of relations reflected in Case and agreement phenomena.