In determining how reliable is reliable enough and how much error can be tolerated in criterion-referenced testing, the following relationships hold: (1) the more specific an objective is, the fewer the items required to reliably measure it; (2) the more specific the objectives are, the more objectives required to cover a given span of the curriculum; and (3) the more specific the objectives are, the more diagnostic the test is in terms of uncovering specific deficiencies in a student's achievement. There is also a four-way interdependency between comprehensiveness, measurement reliability, test length, and curriculum span. Data are presented and discussed with particular emphasis on the reliability problem and specificity-generality dimension of behaviorally stated objectives. Data were collected on the Prescriptive Mathematics Inventory and the Prescriptive Reading Inventory. It is concluded that more information is obtained about a student's specific strengths and weaknesses by measuring a large number of specific objectives with few items than by measuring a few general objectives with a large number of items. (Author/RC)