When it comes to modeling marriage markets, one often considers a setting where anyone on the opposite side of the market can serve as a potential spouse.
In reality, however, we tend to acknowledge that there exist separate (overlapping) markets for the potential couples. In this paper, we aim to learn more about the various factors that define such marriage markets. Using revealed preferences, we develop a method that allows reconstructing (partially) the set of couples that belong to the marriage market, and (exactly) the set of couples that do not belong to it. We apply our results to the data identifying the marriage market for each individual using machine learning techniques, and study its characteristics.