The paper analyzes how and under which assumptions it is possible to compare (in a relationist setting and relatively to qualities) entities living in different worlds. We begin with a standard technique to construct quality kinds via an abstraction process. In the first case, the process is applied across all the possible worlds and we show that the resulting quality system has problematic conseThen, we focus on the alternatives that arise when the abstraction process is applied within each single world independently, i.e., assuming similarity judgments make sense only when referring to entities living in the same world. This situation leads to worlds with unrelated quality systems and we look at the problem of quality comparison across worlds. We analyze under which assumptions this comparison is possible and discuss its limits by considering the structural information that one can infer from the elements shared by (two or more) overlapping worlds. Exploiting the use of such informat...