The development of geographic information systems (GIS) is recognized as a pre-requisite for the effective exploitation of remotely sensed data. Current commercial systems represent a solution with a strong bias from either the mapping or the remote-sensing market. They thus lack full cross discipline functionality and a model-oriented approach. This paper examines some of the key issues in truly integrated GIS design. The five data types of image, object (vector), terrain, tabular and knowledge are identified along with the operations required with them within a GIS. The term `geoschema' is introduced (analogous to schema within a database) to describe the organization of the geographical datasets. Three-dimensional data handling, the necessity of qualifying data and the user interface are given particular attention. An efficient method of implementing an integrated spatial index into the data sets is described.