The gel point of end-linked model networks is determined from computer simulation data. It is shown that the difference between the true gel point conversion, p c , and the ideal mean field prediction for the gel point, p c,id , is a function of the average number of cross-links per pervaded volume of a network strand, P , and thus, contains an explicit dependence on junction functionality f . On the contrary, the amount of intra-molecular reactions at the gel point is independent of f in a first approximation and exhibits a different power law dependence on the overlap number of elastic strands as compared to the gel point delay p c − p c,id . Therefore, p c − p c,id cannot be predicted from intramolecular reactions and vice versa in contrast to a long standing proposal in literature.Instead, the main contribution to p c − p c,id for P > 1 arises from the extra bonds (XB) needed to bridge the gaps between giant molecules separated in space and scales roughly ∝ (P − 1) −1/2 . Further corrections to scaling are due to non-ideal reaction kinetics, composition fluctuations, and incompletely screened excluded volume, which are discussed briefly.