The quality of a geometrical database, especially in urban areas with high real property prices, is an important issue whenever the introduction of an information system is under discussion. The desired high quality usually conflicts with the budget available. Generally, there is a lack of concepts of expressing the given accuracy and of knowledge to identify the required accuracy. The purpose of the paper is to show methods to describe the accuracy of geometric data digitized from existing maps. Under the assumption of independent normally distributed coordinate errors with expectation zero, the distribution of positional errors is derived. The analysis of a sample of 571 digitized points shows: that empirical data contains random and systematic errors, and how this violation of the assumption affects the error distribution. The propagation of positional errors and some possibilities to characterize the accuracy of straight lines and of polygons are discussed.