Software product line modeling has received a great deal of attention for its potential in fostering reuse of software artifacts across development phases. Research on the testing phase, has focused on identifying the potential for reuse of test cases across product line instances. While this offers potential reductions in test development effort for a given product line instance, it does not focus on and leverage the fundamental abstraction that is inherent in software product lines -variability.In this paper, we illustrate how rich software product line modeling notations can be mapped onto an underlying relational model that captures variability in the feasible product line instances. This relational model serves as the semantic basis for defining a family of coverage criteria for testing of a product line. These criteria make it possible to accumulate test coverage information for the product line itself over the course of multiple product line instance development efforts. Cumulative coverage, in turn, enables targeted testing efforts for new product line instances. We describe how combinatorial interaction testing methods can be applied to define test configurations that achieve a desired level of coverage and identify challenges to scaling such methods to large, complex software product lines.