In order to perform accurate Monte Carlo (MC) simulations, which is a stochastic method resulting in uncertainty, variance reduction (VR) techniques are often necessary to reduce the relative error for quantities of interest. The use of weight windows (WWs) is a common VR method in which the statistical weight of particles are changed based on various parameters in the simulation. WWs are most commonly represented as a Cartesian WW mesh (CWWM) where WWs are defined across all energies on each mesh voxel. For large, geometrically complex problems, these meshes often need to be developed with fine resolution over the entire spatial domain in order to capture necessary fine detail in some regions of the geometry. This can cause the memory footprint of these meshes to be extremely large and computationally prohibitive. Furthermore, CWWMs are not necessarily efficient in their implementation with respect to when particle weight is checked and updated. This dissertation work presents a novel method for representing WWs aimed at addressing the computational limitations of CWWMs while also improving VR efficiency. In this method, the WWs are transformed into a faceted mesh geometry, known as a WW isosurface geometry (WWIG), where the surfaces are the isosurfaces derived from the WW values in a CWWM. The WWIGs can then be used during particle tracking with the Direct Accelerated Geometry Monte Carlo (DAGMC) toolkit, which allows for particle tracking on arbitrarily complex geometries. In this work, an algorithm for using WWIGs for MC VR has been implemented in DAGMC coupled with Monte Carlo N- Particle transport code (MCNP) (DAG-MCNP) 6.2. Initial verification and demonstration experiments show that the WWIG method performs accurate and comparable VR to using CWWMs. Further analysis has been done to demonstrate how changing mesh geometric features of the WWIGs affects computational performance during MC radiation transport. Depending on parameters set for generating the WWIGs and the starting CWWM, the isosurfaces of the WWIGs can vary in mesh coarseness, surface roughness, and spacing. In this work, we explore how these different geometric features of the WWIGs affect the memory footprint and computational performance during variance reduction for Monte Carlo radiation transport. In the end, we see that using WWIGs for MC VR improves WW efficiency and is comparable in performance to using CWWMs.