This paper focuses on various aspects of datum definitions and datum design for GPS networks. We consider GPS observations as being a major source for positioning future control points. Hence accuracy, reliability, and sensitivity of the datum definition become important. <br/>Accuracy measures the impact of random errors, and reliability measures the impact of gross errors in coordinate estimation. Sensitivity provides a quality measure for how well point movements are estimated in time. <br/> <br/>It turns out that the accuracy and the reliability of GPS networks are independent of the geometric distribution of the network points; this is in contrast to control networks measured by classical methods. We demonstrate that the distribution of datum points does not affect accuracy and reliability while sensitivity does depend on this distribution. We present a number of rules for the distribution of datum points that improve the sensitivity of the datum. <br/> <br/>The internal reliability does not depend on datum definition. The external reliability, the accuracy and the sensitivity depend on datum definition. It turns out that Baarda's suggestion for measuring the global external reliability is not suitable for GPS networks, another procedure is given.