Radiation therapy relies in great extent on delineations of tumour and organs on medical images. These delineations are essential for the entire treatment. Unfortunately manual delineations are both prone to variation. At the same time the manual delineation process is time-consuming. This thesis represent a work within the automatic definition of organs and tumours. The thesis includes a summary of the prior methods employed for automatic segmentation and 3 articles describing segmentation algorithms of different areas of application for radiation therapy. Variation within and between manual and automatic segmentation methods is documented in the thesis. The last article of the thesis analyses treatment outcome difference due to manual delineation variation.