Statistical procedures employed in item analysis are based on the performance of the individual on a given test. If, however, one desires to assess the efficiency of some treatment, the researcher is interested in the performance of groups of students who received the treatment rather than in the performance of any one individual. In such cases statistics based on individual performance do not constitute a proper basis for item selection and for interpretation of test results. In this context, therefore, the discrimination power of an item should be conceived as its capacity to discriminate among classes and not among individuals. This approach is not only logically sound, but also increases the quantity and quality of the information available. Three new parameters are introduced: item-test correlation based on class data, item standard deviation based on class data, and the intraclass correlation coefficient of an item. An item analysis of a fourth grade arithmetic test is used as an example. Other class data parameters are suggested for future research. (DG)