Primary objective. The aim of the present investigation was to construct and pilot test a computer-based multi-tasking procedure that could be used to assess the ability of patients with neurological damage to remember instructions in the real world. Research design. A simulated street scene was constructed from a network of photographs and sounds that patients could move through using a touch screen. Three patients with severe, moderate or mild cognitive impairment were assessed on a range of neuropsychological tests and three multi-tasking procedures based in the street. The performance of each patient was compared with that of a matched control. Procedure. Three tests were administered, each of which involved 'walking' along the length of the street once. On the first test, participants were given up to five errands to remember while moving along the street. On the other two tests they were given three instructions to carry out repeatedly (e.g. 'Record the name of the nearest shop when you hear a dog bark'). In one condition they were given the three instructions on a sheet they could consult and in the other they had no list. Outcomes and results. In each case the patient performed more poorly on the multi-tasking test than the matched control. The patients' performance on the computer-based tests was consistent with clinical descriptions of their memory-related deficits. Conclusion. The results illustrate the way in which computer presentation of naturalistic stimuli can be used to construct flexible and standardized tests of memory functioning that have enhanced ecological validity.