System testing in the automotive industry is a very expensive and time-consuming task of growing importance, because embedded systems in the domain are distributed over numerous controllers (ECUs). Modern software development techniques such as continuous integration require regular, repeated and fast testing. To achieve this in the automotive domain, test suites for a specific software change must be tailored. We propose a novel test selection technique for system-level functions in the automotive industry based on component and communication models. The idea is to follow input and output signals that are used in the testing steps through the ECUs implementing a function. We select only those tests for a planned integration in which at least one of the signals sent in its steps is processed by the ECU that was changed and thus triggered the integration. The technique is well-suited for black-box testing since it requires only the full test suite specification and the system architecture. We applied the technique to a test suite of the Active Cruise Control function at BMW Group in the context of hardware-in-the-loop system testing and found the possible reduction rates to be 82.3% on average in comparison to the full test suite. Possible future work includes the evaluation with a wider set of functions, the evaluation of the fault detection rate, further automation and combination with other test selection techniques.