In this article, I argue for an ellipsis analysis of Japanese null arguments on the basis of a novel observation that covert extraction (i.e., extraction that does not affect word order) is possible out of them. Specifically, assuming that the extraction possibility is a diagnostic for surface anaphora/ellipsis, I claim that the covert extraction possibility indicates that Japanese null arguments can be elliptic: they cannot be uniformly silent deep anaphora/proforms. Furthermore, I show that there is an overt/covert extraction asymmetry in that only covert extraction is allowed out of Japanese null arguments. I argue that the LF copy analysis of argument ellipsis provides a solution for the overt/covert extraction asymmetry. The discussion also has consequences for the proper analysis of several phenomena of Japanese syntax, including wh-in-situ.