Direct assessment of writing skill, usually considered to be synonymous with assessment by means of writing samples, is reviewed in terms of its history and with respect to evidence of its reliability and validity. Reliability is examined as it is influenced by reader inconsistency, domain sampling, and other sources of error. Validity evidence is presented, which shows reported relationships between direct assessment scores and criteria such as class rank, English course grades, and instructors' ratings of writing ability. Evidence on the incremental validity of direct assessment over and above other available measures is also given. It is concluded that direct assessment makes a contribution but that methods need to be developed to improve its reliability and reduce its costs. New automated methods of textual analysis and new kinds of direct assessment in which more than a single score is produced are suggested as two approaches to better direct assessment. (Author)