A central element in military expert-system (es) application is the dynamic handling and adaptive control of evidence and uncertainty. There are currently five distinct approaches to the problem of evidential reasoning (er). These are the bayesian, neyman-pearson, dempster-shafter, kyburg, and possibility/fuzzy approaches. In addition to the er task of revising degrees of belief or support, we must control the accumulation of evidence. The er approaches do not currently address questions of evidential control. However, the fields of sequential statistics, sequential decision analysis, and sequential pattern recognition provide several candidate control techniques. The proposed effort will first apply current sequential control techniques to each of the er approaches, thus clearly identifying incompatibilities between these techniques and the er theories. It will then analyze these incompatibilities, thus providing a framework for, and initiating development of, control techniques triat can be better interfaced with each of the er methods for military expert systems.