Home range analysis involves characterizing the spatial extent that an animal occupies from sample points that record its location periodically over time. Kernel density estimation (KDE) is currently the most widely applied and accepted method of home range estimation, although several authors have recently questioned its use for this purpose, citing instances when it performed poorly for certain types of point distributions. The first part of this dissertation provides a critical evaluation of KDE in the context of home range estimation from a geographic information science (GIScience) perspective. First, the accuracy of KDE as a home range estimator is tested using simulated animal locational data that conform to different shapes. Because those results suggest that KDE is not robust to point pattern shape, the method then is examined in the context of its underlying statistical and spatial assumptions. This review reveals that KDE implicitly assumes that the point locations used in the analysis were generated by a stationary, Euclidean-based process. As point locations for home range analysis are derived from an animal’s continuous movement trajectory through space, a nonstationary, network-based process, application of KDE to home range analysis is in violation of the technique’s underlying assumptions. This leads to the conclusion that KDE is inappropriate for home range estimation. The second part of this dissertation then develops and explores an alternative method of density estimation that assumes network-based rather than Euclidean-based space usage: network-based kernel density estimation (NKDE). NKDE is applied to wildlife-vehicle collision data for illustration. Because animal locational data are generated by a network based process, NKDE is extended to estimate wildlife home ranges. Then, NKDE is applied to the same point pattern data of different shapes used to evaluate KDE. The results suggest that NKDE performs much more accurately as a home range estimator than traditional KDE.