Scenes from nature share in common certain statistical properties. Images with these properties can be processed efficiently by the human brain. Patterns with unnatural statistical properties are uncomfortable to look at, and are processed inefficiently, according to computational models of the visual cortex. Consistent with such putative computational inefficiency, uncomfortable images have been demonstrated to elicit a large haemodynamic response in the visual cortex, particularly so in individuals who are predisposed to discomfort. In a succession of five small-scale studies, we show that these considerations may be important in the design of the modern urban environment. In two studies we show that images from the urban environment are uncomfortable to the extent that their statistical properties depart from those of scenes from nature. In a third study we measure the haemodynamic response to images of buildings computed as having unnatural or natural statistical properties, and show that in posterior brain regions the images with unnatural statistical properties (often judged uncomfortable) elicit a haemodynamic response that is larger than for images with more natural properties. In two further studies we show that judgments of discomfort from real scenes (both shrubbery and buildings) are similar to those from images of the scenes. We conclude that the unnatural scenes in the modern urban environment are sometimes uncomfortable and place excessive demands on the neural computation involved in vision, with consequences for brain metabolism, and possibly also for health.