In this paper, we proposed a new guidance scheme of the GPS-GIS-based speech guidance system using the adaptive fuzzy inference neural network (AFINN). Geographic Information System (GIS) calculates some attributes of user-to-landmark relations, so-called, landmark celebrity, landmark facade, landmark area, representative point of latitude and longitude (the center of gravity) of the landmark, user's latitude and longitude from GPS, and user's direction. These attributes were transformed to six basic elements, such as the celebrity, the facade relative bearing, the Euclidean distance between the landmark and user, the area size, the direction relative bearing and the user's individual data, respectively. The AFINN, that has the above six elements as the input data, calculates a level of importance (LOI) corresponding to a "cognitive distance". We considered that acceptable guidance was achieved by using the LOI. We applied the analytic hierarchy process (AHP) to the leaning data of the AFINN. Therefore, the guidance system reflected user's preference, and obtained the guidance suitable for diverse users. Numerical simulations suggested that the AFINN should be available for determining the guidance satisfying each user. Finally, the visually impaired pedestrians evaluated the guidance system from the user's viewpoints.