The research described in this paper examines the relationship between raters and the training raters receive to apply that rating. The primary concern is maintaining scoring consistency within the context of raters' perceptions. If the criteria for holistic and/or analytic scoring can be recognized as communication based, rather than topic based, then student essays could be analyzed as features or qualities registered by the raters who read those essays. This paper examines 300 raters' perceptions of quality writing before training, after training, and following the reading of 90,000 student essays. Analyses of the holistic raters' perceptions of writing quality on a 1 to 6 score scale indicate 20 characteristics, grouped into six general constructs: (1) content; (2) organization; (3) sentence construction; (4) usage; (5) mechanical characteristics; and (6) textual characteristics. For pre-training scores the relationships among constructs remains consistent across ability levels. However, following training differentiation exists across all six categories, particularly at the extremes of the scales. Raters still seem less able to develop patterns of relationships among the essay characteristics at the middle points of the ability scales, even following training and reading of essays. (PN)