Human-autonomy teaming using physiological sensors poses a novel sensor fusion problem due to the dynamic nature of the sensor models and the difficulty of modeling their temporal and inter-subject variability. Developing analytical models therefore requires defining objective criteria for selection and weighting of sensors under an appropriate fusion paradigm. We investigate a selection methodology grounded in two intuitions: 1) that maximizing the relevance between sensors and target classes will enhance overall performance within a given fusion scheme; and 2) that minimizing redundancy amongst the selected sensors will not harm fusion performance and may improve precision and recall. We apply these intuitions to a human-autonomy image classification task. Preliminary results indicate strong support for the relevance hypothesis and weaker effects for the redundancy hypothesis. This relationship and its application to human-autonomy sensor fusion are explored within a framework employing three common fusion methodologies: Naive Bayes fusion, Dempster-Shafer theory, and Dynamic Belief Fusion.