Content Based Image Retrieval (CBIR) is a significant research area. The challenge for CBIR researchers is to develop techniques for analysing the low level content of images, and then representing that content in a way that can be effectively searched and compared. Low level content of images comprises of information about shape, colour and texture. The focus of this thesis is shape descriptors. Shape descriptors are categorised as spatial or spectral methods. Another common approach for categorisation of shape descriptors is region-based or contour-based methods. Spatial characteristics preserving shape descriptors are proposed in this thesis. Spatial characteristics are structural features of shape. Two shapes which are structurally similar have similar spatial characteristics. New descriptors proposed in this thesis are called Spherical Harmonics Descriptors. These are spectral descriptors based on spherical harmonics. These descriptors implicitly represent shape spatial characteristics. The proposed descriptors are generic i.e. they can be adopted for shape region and shape contour. The proposed approach is compared with Generic Fourier Descriptor (GFD) because GFD has been shown to be amongst the most effective shape descriptors. The proposed approach performs better than GFD by overcoming the drawbacks of GFD. Experimental results also show that the proposed approach is more effective than the MPEG-7 descriptors for shape region and shape contour, namely, Zemike Moments Descriptor and Curvature Scale Space respectively. Spherical Harmonics Descriptors are proposed based on extensive investigation of existing descriptors. Existing descriptors, in the spatial domain and the spectral domain, are investigated. Distance histogram, angular histogram and intensity histogram are investigated in the spatial domain. GFD is investigated in the spectral domain. It is found that the discriminating ability of these descriptors is potentially enhanced by complementing them with proposed spatial characteristics of shape.