A new approach to item analysis, concerned not only with differences among individuals, but also differences among the groups in which the individuals may have membership, is described. This multi-level aspect may be applied to studies of differences among individuals, classes, and schools, depending upon the kind of problem being studied and the manner in which the sample is obtained. Most educational data are multi-level; data are collected from individuals (students, teachers) within groups (classrooms, schools). Observed group effects may be the result of the ways in which individuals are selected into the groups or of the effects on individuals of the programs and processes which take place within the groups. Empirical item analysis usually combines information from students in multiple groups, resulting in reliability and validity coefficients which combine between and within group relationships. Application of multi-level regression techniques to item analysis is presented, using biology test data for the United States from the International Association for the Evaluation of Educational Achievement Six-Subject Survey. Statistical procedures for disaggregating the between and within group relationships are presented, and possible interpretation of the results and implications for program evaluation are discussed. (Author/CTM)