Two specific methods of analysis in large-scale evaluations are considered: structural equation modeling and selection modeling/analysis of non-equivalent control group designs. Their utility in large-scale educational program evaluation is discussed. The examination of these methodological developments indicates how people (evaluators, methodologists, agency staff) involved in the design and conduct of large-scale program evaluation might approach decisions concerning appropriate methodology and its proper use. Evaluation activities and the range of methodological issues considered are: field-based investigations of large-scale programs; evaluation of on-going programs and of various forms of social experiments; and both well-defined and broad-based educational programs. Both analytical procedures employ explicit models of the phenomena believed to be responsible for the difficulties in estimating program effects. Both are also adaptable to situations where there are no specific comparison or control groups and where panel data exists on program participants. A discussion of the LISREL model and its limitations as an analytical approach to estimation in structural equation modeling with latent variables is included. (PN)