A classical model for achievement test development was influenced by: the educational tradition of selecting a sample of questions appropriate to a subject and counting the proportion of acceptable responses; the emphasis on obtaining a large and homogeneous sample of questions to insure reliability; and the tradition in psychological testing of scaling performance to meaningful reference groups. The steps in developing tests are: planning; item writing/review; item tryout; final assembly; and norming/scaling. The classical model offers suggestions for selecting tests and evaluating programs for disadvantaged or minority students. Evaluation should extend over two or three years and measure the impact of anticipated achievement gains on self concept and school attitudes. Evidence from the literature or from experts familiar with the local context should indicate that the test is appropriate for program participants, and that it has been screened for bias. To avoid discouraging slow learners or boring high achievers, out-of-level testing is appropriate, but a careful check on the scaling procedure is crucial. To solve the problem of comparable norms, existing long-term data for a school should be used. Standardized test results should not be ignored if they fail to meet expectations; the data require the more difficult task of interpretations. (CP)