the first relies on the knowledge of the 3D structure of your target, the second on the information coming from known ligands binding to the target protein (i.e. 2D chemical similarity analysis of known ligands)
Ligand-based design uses only knowledge about the ligands, for example in cases where the structure of the target protein is not known. Typical applications are pharmacophores and QSAR. Structure-based design on the other hand require a suitable 3D structure of the target, either from X-ray crystallography or homology modeling in order to do docking.