When we have access to more than one source of information about a decision we have to make, it is often difficult to know how to fuse that information so as to determine our best decision. A quantitative approach to identification fusion is presented. With no prior knowledge of the ‘goodness’ of each source’s performance, one is reduced to utilizing a voting scheme. The approach presented in this talk assumes prior knowledge of the ‘goodness’ of each source in the form of the Confusion Matrix for each source for the problem at hand. Confusion Matrices are introduced and discussed. Then a process for optimally fusing two or more information sources, given that we know their Confusion Matrix characterizations, is presented. The technique is demonstrated utilizing sets of radar and FLIR (forward looking Infrared) decisions as to the type (e.g., tank, truck, jeep, etc.) when observing a set of military vehicles. It is shown that fusing the individual decisions in this way provides better identification performance than either the radar or FLIR sensors alone provide. Performance improvements are quantified and it is further shown that this technique can be tailored so as to function at any of a number of algorithm operating points and that the fusion algorithm is relatively robust to errors in the single-source Confusion Matrix characterizations.