In this paper we describe how to infer state machine models of systems from legacy unit test suites, and how to generate new tests from those models. The novelty of our approach is to combine control dependencies and data dependencies in the same model, in contrast to most other work in this area. Combining both kinds of dependency helps us to build more expressive models, which in turn allows us to produce smarter tests. We illustrate those techniques with examples from our implementation, the James tool, designed to apply these techniques in practice to Java code and tests.