Finding a least-cost-path in a raster data format is a useful function in geographical information systems. However, existing algorithms are often inadequate for practical roadway planning. This paper improves conventional algorithms by including the considerations of spatial distances, anisotropic costs and the presence of bridges and tunnels in the paths. This new algorithm is implemented in JAVA to run with actual remote sensing and DEM data. The experimental results show that this approach produces realistic least-cost paths for practical roadway planning.