Abstract A data parallel algorithm is presented for spatial data (raster-to-vector) conversion. It operates on pure two-dimensional raster images and derives a fully topological vector data set, where polygons are defined by homogeneity criteria. The algorithm is implemented in data parallel C, and makes extensive use of scan functions that combine communication and computation along grid axes. Code segments listing each step in building a hierarchy of points, lines, and polygons are listed. Performance evaluations on a Connection Machine-2 reveal that run-time is not sensitive to the number of polygons in the image, and increases sublinearly with grid size.