Using a base of data developed from the December 1979 administration of the College Board's English Composition Test, several linear models were developed to predict judgments of writing samples made by English professors. Other models were hypothesized through confirmatory factor analysis. All models and factors were related to two external criteria, high school English grades and self-reported writing ability. The most valid models were those based on focussed ratings of three key aspects of the writing samples: discourse quality, syntactic quality, and lexical quality. The least valid models were those based on checklist ratings of more atomistic qualities such as organization, tone, and sentence variety. The best fitting-factor model was a four-factor model with Verbal Aptitude, English Achievement, Mechanics Skills, and Discourse Skills as the factors. Factor extensions on the two external criteria showed reasonable consistency across four populations sampled. The economics of writing assessment was also examined.