Pervasive context-aware software is an emerging kind of application. Many of these systems register parts of their context-aware logic in the middleware. On the other hand, most conventional testing techniques do not consider such kind of application logic. This paper proposes a novel family of testing criteria to measure the comprehensiveness of their test sets. It stems from context-aware data flow information. Firstly, it studies the evolution of contexts, which are environmental information relevant to an application program. It then proposes context-aware data flow associations and testing criteria. Corresponding algorithms are given. It uses a prototype testing tool to conduct experimentation on an RFID-based location sensing software running on top of context-aware middleware. The experimental results show that our approach is applicable, effective, and promising.