Two empirical studies are presented which seek to extend the parallels between disordered speech production in aphasia and in normals. Study 1 compares the rate and distribution of some theoretically interesting error types in a jargon aphasic and a normal error corpus. Study 2 is an investigation of how the error pattern of normal speakers evolves as utterances become more practiced. On the basis of these studies, we offer a hypothesis about the nature of the variation between more and less disordered systems. Our claim, which is developed in the context of spreading activation models of production, is that such variation is tied to the ability of the system to deliver activation to intended units, relative to that of unintended units, within the time required by the task at hand.