The distribution of Abell clusters of galaxies is analysed to study the regularity of the supercluster-void network. A new geometric method sensitive to the regularity of the location of clusters is applied. We find that the supercluster-void network resembles a cubical lattice over the whole space investigated. The distribution of rich superclusters is not isotropic: along the main axis of the network it is periodic with a step of length about 130 Mpc (for Hubble constant h=1), whereas along the diagonal of the network the period is larger. This large-scale inhomogeneity is compatible with recent CMB data.

Recently, the observed cellular nature of the large-scale structure of the Universe with its quasi-regular pattern of superclusters and voids has been pointed out by several authors. In this paper, we investigate properties of the initial power spectrum which lead to prediction of structure consistent with these observations.For this purpose, we analyze the evolution of structure within four sets of 2- and 3-dimensional cosmological models, which differ in their initial power spectrum. The models include HDM and CDM models as well as double power-law models. We discuss in detail the impact of model parameters such as the large scale and small scale power and the position and height of the maxima of the power spectra on the predicted structure. The best agreement with observations was observed in the model with the Harrison-Zeldovich spectrum on large scales, a power index $n\approx -1.5$ on small scales, and a maximum of the power spectrum at $\approx 130$~\Mpc. In this model the distribution of masses of clusters and superclusters, the correlation function of clusters, and the void distribution reproduce well the respective observed distributions. We also investigated the dynamical evolution of the supercluster-void structure. The results show that the basic supercluster-void network is formed very early and is essentially given by initial conditions.