A new method for increasing confidence in software based on the premise that competent programmers write correct or 'nearly' correct software is presented. The envisioned system takes as input a program and a set of test data. It produces and executes a set of perturbation programs, and generates a list indicating which perturbation programs are indistinguishable from the original program (with the given data). A non-empty list indicates that the data is not adequate, that there exist equivalent programs in the list, or that the original program is incorrect. An empty list indicates that the original program is either correct or 'far' from correct. While the set of perturbation programs should be large enough to include many commonly made errors, it appears that there is a coupling effect suggesting that errors not present in the set of perturbation programs are still checked by this method. Two examples of the use of this method are given. (Author).