Accurate, high-resolution 3D mapping of environmental terrain is critical in a range of disciplines. In this study, we develop a new technique, called the PCFilt-94 algorithm, to extract 3D point clouds from coarse resolution millimetre-wave radar data cubes and quantify their associated uncertainties. A technique to non-coherently average neighbouring waveforms surrounding each AVTIS2 range profile was developed in order to reduce speckle and was found to reduce point cloud uncertainty by 13% at long range and 20% at short range. Further, a Voronoi-based point cloud outlier removal algorithm was implemented which iteratively removes outliers in a point cloud until the process converges to the removal of 0 points. Taken together, the new processing methodology produces a stable point cloud, which means that: 1) it is repeatable even when using different point cloud extraction and filtering parameter values during pre-processing, and 2) is less sensitive to over-filtering through the point cloud processing workflow. Using an optimal number of Ground Control Points (GCPs) for georeferencing, which was determined to be 3 at close range (<1.5 km) and 5 at long range (>3 km), point cloud uncertainty was estimated to be approximately 1.5 m at 1.5 km to 3 m at 3 km and followed a Lorentzian distribution. These uncertainties are smaller than those reported for other close-range radar systems used for terrain mapping. The results of this study should be used as a benchmark for future application of millimetre-wave radar systems for 3D terrain mapping.