This paper provides an empirically more adequate account of Japanese coordination than Johannessen (1993/1996), Zoerner (1995), Yi (1997/2000). Assuming that the putative conjunction to is not a head of coordination but a mere conjunct marker, our analysis allows for the head-initial constituency in sosite-coordination in Japanese. Our analysis further suggests that to-coordination without an overt conjunction sosite involves a null conjunction. It follws from this analysis that (i) a single conjunct marker to cannot conjoin multiple conjuncts, (ii) by contrast, a single 'true' conjunction, i.e., sosite, can conjoin multiple conjuncts, and (iii) the initial conjunct, assuming Hoji (1998)，asymmetrically c-commands a non-initial conjunct.