Variable interaction is an important aspect of a problem, which reflects its structure, and has implications on the design of efficient optimization algorithms. Although variable interaction has been widely studied in the global optimization community, it has rarely been explored in the multi-objective optimization literature. In this paper, we empirically and analytically study the variable interaction structures of some popular multi-objective benchmark problems. Our study uncovers nontrivial variable interaction structures for the ZDT and DTLZ benchmark problems which were thought to be either separable or non-separable.