The use of confirmatory factor analytic procedures to examine the dimensionality of writing skills as measured by a large-scale direct writing test was illustrated. Internal construct validity evidence about the nature of writing skills measured by the test was provided. Data used were scores assigned by about 100 trained professional raters on a state-wide competency-based test of writing skills. A total of 3,430 students responded to all four prompts of the writing tests. A sequential hypothesis testing strategy was followed for a one-factor model, a two-factor mode-specific model, and a two-factor occasion-specific model. These were compared to a null model specifying no common factors. Results from these analyses provided evidence about the usefulness of the construct-based hypotheses for explaining the variability in data from this test. A one-factor model provided a more plausible and parsimonious representation of the data than did the two-factor hypotheses. Evidence in support of a unidimensional interpretation of the scores suggests there may be little practical distinction between narrative and explanatory scores. A table displays goodness-of-fit statistics. (SLD)