The proposed theoretical approach views the criminal justice system as an interconnected network of decision 'nodes,' with a given node consisting of a class of decisionmakers required by procedure or law to reach a specified decision. A connection between two nodes is assumed to exist if events in one node control the decisions reached in the other node. The first step in constructing a predictive theory of the criminal justice system based on the proposed theoretical approach is to define the set of decisions and classes of participants making those decisions and to place these decision nodes in the appropriate temporal order. The next step in developing a predictive model is to select the unit of analysis that is to be used in constructing empirically derived decision rules for the various nodes. The empirical derivation of models of the decision strategies used in each node is the last step in the process. The goal of this stage of the analysis is to select from the many possible factors (i.e., categories of information and sources of influence) those that account for the largest percentage of variance in the decision being studied. This approach permits the evaluation of claims that certain factors have not been properly considered by decisionmakers. Forty-six references are listed.