Modem spatial information systems create precise digital representations of features that are often
inherently fuzzy, inaccurately located, or otherwise subject to a variety of forms of uncertainty.
In this environment it is all too easy to impute levels of accuracy to the results of analysis and
processing that are wildly inconsistent with the truth. The paper begins with a series of examples
of the uncertainty present in map data, and the consequences of its use in GIS. In many cases,
the map acts as a channel of communication between the scientist or surveyor in the field, and the
eventual user of geographic data; uncertainty can be interpreted from this perspective as the
result of inefficiency in the communication channel, providing the user of the data with only a
small fraction of the knowledge available to the observer or interpreter in the field. Two
approaches are proposed to deal with this problem. The first requires a change of methodology, and
makes the database the communication channel rather than the map--cartographic representations are
seen in this approach as views of the database with their own objectives and priorities. The
second adopts a statistical perspective by modeling uncertainty through the construction of error
models. These are illustrated, and an example of the use of an error model to estimate error in the
measurement of area in a GIS is discussed. The paper ends with a discussion of alternative
methods for visualization of spatial accuracy within the context of error modeling.