An optimally designed geodetic network is characterised by an appropriate level of precision and the lowest possible setup cost. Reliability, translating into the ability to detect blunders in the observations and higher certainty of the obtained point positions, is an important network characteristic. The principal way to provide appropriate network reliability is to acquire a suitably large number of redundant observations. This approach, however, faces limitations resulting from the extra cost. This paper analyses the possibility of providing appropriate reliability parameters for networks with moderate redundancy. A common problem in such cases are dependencies between observations preventing the acquisition of the required reliability index for each of the individual observation. The authors propose a methodology to analyse dependencies between observations aiming to determine the possibility of acquiring the optimal reliability indices for each individual observation or groups of observations. The suggested network structure analysis procedures were illustrated with numerical examples.