The aim of this paper is to shed new light on collocational analysis, reviewing the main stream of research on this issue and trying to overcome some of its intrinsic problems, such as determining the optimal span and partially explaining the reason for undesired collocates (statistically significant collocates, though lexically and semantically not related to the node word). The idea is not just to calculate significant collocates of a chosen node word or to elucidate which is the best statistical procedure to achieve this goal. The aim is to investigate the way words socialise with other words, forming complex network-like structures or units: lexical constellations. This behaviour cannot be explained solely on a grammatical and/or semantic basis or even based on the present state of the art in collocation research.