Over the past decade, the human micro-Doppler signature has been a subject of intense research. In particular, much work has been done in relation to computing features for use in a variety of classification problems, such as arm swing detection, activity classification, and target identification. Although dozens of features have been proposed for these purposes, little work has examined the issue of which features are more important - i.e., have a greater impact on classification performance - than others. In this work, an information theoretic approach is applied to compute the importance ranking of features prior to classification for the specific problem of discriminating human walking from running. Results show that the ranking of features according to mutual information directly relates to classification performance using support vector machines.