The availability of choice is a neglected aspect in studies of geographical accessibility, which typically concentrate on distance to the nearest service. Records of patient registrations with general medical practices offer the opportunity to examine the geographical distribution of choice for an essential service. This population study of two million residents of Cambridgeshire, Norfolk, and Suffolk used postcodes extracted from patient registers and estimated car travel times from residential locations to general practice surgeries in a geographical information system. Only 56% of the population were registered with the practice nearest their home. People were more likely to use the nearest practice if they lived in rural rather than urban areas and where a surgery was within walking distance. Choice, as measured by the number of practices used by 95% of residents, was highest in the larger urban areas and lowest in small towns and rural areas with a local surgery. Ten percent of the population were served by monopoly practices. Overall, the distribution of registrations reflected a regular and predictable substitution of choice for increased travel time. People were 29% less likely to register with a practice for every additional minute of travel time: an almost perfect distance-decay relationship that was used to calibrate a potential model of accessibility over the study area. The resulting values of potential accessibility were found to approximate the combination of travel time to the nearest surgery and the actual range of choice exercised by residents. This demonstration that the potential accessibility model reflects consumer behaviour has applications beyond the health field.