A method is described for choosing sample sizes when the domain to be sampled consists of a finite set of sentences and the purpose is to construct a test to assess the comprehension or the readability of written discourse. The testing method is that proposed in Bormuth's work on transformational analysis within a criterion-referenced measurement framework. The suggested sampling unit for this type of analysis is the sentence, but other units are mentioned. Guidelines are provided for computing sample size based on: the size of the domain, the criterion level for successful performance, and the confidence level desired. Two methods are suggested for selecting the sample of sentences to which the transformational rules for item writing are to be applied. A table of sample sizes for a given set of conditions is presented to aid the practitioner in using this sampling procedure. (Author/CTM)