Vertical stereographic aerial photographs obtained as photographic plates or, more recently, as digital images represent a primary source of data for desk studies on all engineering projects. New methods of data acquisition from both airborne and spacecraft-based scanners provide unparalleled opportunities to obtain high-resolution data. However, interpretation of the images remains as much an art as a science and is critically dependent on the skill and experience of the interpreter. Relevant experience can be obtained only by the practice of ground-truthing an interpretation that the interpreter has carried out. Although it has rarely been quantified there are important lessons for all engineering projects on the divergence between the interpreted image and the actual ground situation. Some of these are exemplified in a case study undertaken by the authors using large-scale aerial photographs for a landslide hazard investigation in Spain. Based on this investigation it can be assumed that up to 50% of landslide features of engineering geological significance might not have been identified by the aerial photograph interpretation. The alternative is to use the images to divide the landscape into terrain units that allow the landslide potential to be evaluated. It can be demonstrated that this approach is more effective than trying to create a landslide inventory based on aerial photographs alone. Given the potential expansion in the use of large-scale image analysis associated with the recent improvement in data acquisition systems this has implications for the use of images and the training of interpreters.