This paper discusses the use of relationships within data subsets, as viewed in marginal scatterplots, for finding and interpreting patterns in multivariate data. The data are represented by juxtaposition of scatterplots corresponding to variable pairs or triples. Subsets are defined either graphically or by the use of relational and Boolean operations on sets. Points in the same subset are linked across plots by color and symbol. The paper discusses strengths and weaknesses of exploration based on subsets linked across marginal scatterplots. Extensions to three-dimensional scatterplots are briefly indicated.