Initiation of adenovirus DNA replication occurs by a jumping back mechanism in which the precursor terminal priming protein (pTP) forms a pTP.trinucleotide complex (pTP.CAT) catalyzed by the viral DNA polymerase (pol). This covalent complex subsequently jumps back 3 bases to permit the start of chain elongation. Before initiation, pTP and pol form a tight heterodimer. We investigated the fate of this pTP.pol complex during the various steps in replication. Employing in vitro initiation and elongation on both natural viral templates and synthetic oligonucleotides followed by glycerol gradient separation of the reaction products, we established that pTP and pol are separated during elongation. Whereas pTP.C and pTP. CA were still bound to the polymerase, after the formation of pTP. CAT 60% of the pTP.pol complex had dissociated. Dissociation coincides with a change in sensitivity to inhibitors and in Km for dNTPs, suggesting a conformational change in the polymerase, both in the active site and in the pTP interaction domain. In agreement with this, the polymerase becomes a more efficient enzyme after release of the pTP primer. We also investigated whether the synthesis of a pTP initiation intermediate is confined to three nucleotides. Employing synthetic oligonucleotide templates with a sequence repeat of two nucleotides (GAGAGAGA ... instead of the natural GTAGTA ... ) we show that G5 rather than G3 is used to start, leading to a pTP. tetranucleotide (CTCT) intermediate that subsequently jumps back. This indicates flexibility in the use of the start site with a preference for the synthesis of three or four nucleotides during initiation rather than two.