The paper addresses the issue of comparing the reliability of two standard types of networks: serialparallel and parallel-serial. Four variants of dynamic mathematical models are analyzed depending on the lifetime cumulative distribution function of each units of the network, the non-random / random character of the number of units in each subnet and of the number of subnets. Sufficient conditions have been determined for serialparallel networks to be more reliable than parallel-serial networks. The main result is that these conditions do not imply the lifetime distribution of each unit but only the probabilistic distribution of the numbers of units and subsystems of the networks.