It is widely believed that the continuous random network (CRN) model represents the structural topology of amorphous silicon. The key evidence is that the model can reproduce well experimental reduced density functions (RDFs) obtained by diffraction. By using a combination of electron diffraction and fluctuation electron microscopy (FEM) variance data as experimental constraints in a structural relaxation procedure, we show that the CRN is not unique in matching the experimental RDF. We find that inhomogeneous paracrystalline structures containing local cubic ordering at the 10 to 20 angstrom length scale are also fully consistent with the RDF data. Crucially, they also matched the FEM variance data, unlike the CRN model. The paracrystalline model has implications for understanding phase transformation processes in various materials that extend beyond amorphous silicon.