We present a dynamic partitioning strategy that selects test cases using online feedback information. The presented strategy differs from conventional approaches. Firstly, the partitioning is carried out online rather than off-line. Secondly, the partitioning is not based on program code or specifications; instead, it is simply based on the fail or pass information of previously executed test cases and, hence, can be implemented in the absence of the source code or specification of the program under test. The cost-effectiveness of the proposed strategy has been empirically investigated with three programs, namely SPACE, SED, and GREP. The results show that the proposed strategy achieves a significant saving in terms of total number of test cases executed to detect all faults.