The spinal cord is an elongated nervous structure that together with the brain forms the central nervous system. It relays sensory and motor information between the brain and the body, thus controlling most somatic and autonomic body functions. In recent years, great progress has been made in creating digital atlases for the mouse brain, covering regions, cell composition and connectivity. For the spinal cord, however, such atlases do not yet exist. In this dissertation, I present the first versions of high-resolution 3D atlases of the spinal cords for five mammalian species: mouse, rat, marmoset, rhesus monkey, and human. First, I summarize the current state of the efforts to understand the spinal cord and its organization principles from anatomy (cytoarchitecture) to cell classifications and connectome principles. I then describe how the different properties of the spinal cord can help us in reconstructing the various spinal structures. Next, I present a workflow for data-driven reconstruction that can be used to create 3D models of the spinal cord of any species based on available annotated 2D microscopy images. I used data from Sengul, Watson, et al., 2012 to generate 3D reference volumes of the cytoarchitecture of the 5 mammalian species: mouse, rat, marmoset, rhesus monkey and human. Every 3D reference volume model comprises six levels of the spinal cytoarchitecture and defines their locations, shapes, and volumes. This allows us to compare the spinal cord of these 5 species. I find that all reconstructed spinal cords are remarkably similar, showing far fewer differences than the brains of the respective species. For the mouse spinal cord, I use data from the Allen Institute for Brain Science to generate a cell atlas for the three major spinal cell types: interneurons, motoneurons, and glial cells. The mouse spinal cell atlas comprises the densities and positions of 11,7 million cells in total with 2.7 million interneurons, about 31 thousand motoneurons, and 9.0 million glial cells. Next, I discuss how recent transcriptomic data can be used to assign a gene expression profile to each neuron in the spinal cell atlas. I present a preliminary analysis of RNA-sequencing and in situ RNA hybridization data aiming to develop a workflow for labeling of spinal cells with appropriate transcriptome profiles. Finally, I summarize the results of my dissertation as well as consequences and limitations of the underlying assumptions. This discussion ends with an outlook for how the atlas can be developed further in the future.