This chapter describes how segmentation and registration can be used to synthesise medical images of a particular modality from images of another modality. Segmentation-based approaches can generally be decomposed into two components: the first one consists in segmenting the source image and the second one consists in assigning intensity values to the different tissue classes obtained to generate the desired image. The segmentation can be manual or automatic and the intensities can be assigned in bulk (i.e. with predefined values assigned to each tissue class) or in a subject-specific manner. In registration-based methods, an atlas is deformed to match the subject's anatomy using non-rigid registration. The atlas can be composed of a single image of the target modality, of a pair of images from the source and target modalities, or of multiple pairs. Both the general principles and particular examples of segmentation-based and registration-based image synthesis approaches are described. The chapter takes as guiding thread the synthesis of computed tomography from magnetic resonance (MR) images, which is the predominant synthesis task as it answers two very concrete applications: attenuation correction of positron emission tomography (PET) data, particularly for PET/MR scanners, and radiotherapy treatment planning from MRI only.