Studies of aging offer special methodological challenges to the researcher in that he must often examine the change of multiple correlated variables over time. We present a set of procedures that are specifically designed to model change in such multivariate situations. These procedures, which we will call Markov Network Analysis, are directly applicable to modeling change from longitudinal or serial data. In such cases, the parameters of the model have dynamic interpretations, e.g., as coefficients in positive or negative feedback loops. In cross-sectional data, one cannot directly estimate the dynamic coefficients but the model does show how certain dynamic interpretations can be made. Statistically, maximum likelihood estimation procedures are developed and presented. In the development of the statistical model, it is shown how the bias of sequential hypothesis testing, a frequent occurrence in the estimation of complex covariance structure models, may be reduced.