Functional MRI ( fMRI) shows areas of the brain that are active during a task, but the standard approach ( offline analysis after the imaging has finished) precludes tailoring of the imaging to the individual patient, e. g. for assessing normal function around an individual lesion. The aims of the study were to explore the technical feasibility of acquiring functional images in real-time (rt-fMRI), develop the necessary software interfaces and protocols for image acquisition, and to compare images of functional activation acquired in real-time with the standard offline statistical parametric method in patients with solitary brain tumours. Patients with a solitary supratentorial lesion were studied. The rt-fMRI paradigms were sequential finger opposition, ankle movement and language function ( correct recognition of grammatically violated sentences). Datasets were analysed using AFNI software ( National Institute of Mental Health, Bethesda, Maryland, USA) for the real-time analysis and SPM99 ( Functional Imaging Laboratory, University College, London, UK) for the offline analysis. From 11 patients, useful data were obtained in nine. The finger tapping task produced most consistent activation between real-time and offline analysis with good anatomic localization to the primary motor cortex contralateral to the tapping finger. Ankle movement produced weaker activation and correlation with real-time analysis. For the language task the offline analysis provided reproducible activation patterns, but the real-time method showed no activation at the chosen threshold of p=0.001. Tumourous areas of brain did not show any activation with either method of analysis during any task. rt-fMRI is feasible and could be a valuable functional evaluation tool in the planning of surgery for tumours in motor regions of the brain. Further paradigm development is required for evaluation of language, and possibly other more complex executive functions.