Existing cognitive science and psychology studies suggest that a bi-level approach to fault localization is needed with both shallow and deep reasoning. This approach form the underpinnings for developing our Conceptual Model for Software Fault Localization (CMSFL) to aid programmers with the problem of software fault localization. Our CMSFL proposes that, during the fault localization process programmers build two mental models: an actual code model (the buggy code), and an expectation model (the correct code). A multi dimensional approach is suggested with both shallow and deep reasoning phases to enhance the probability of localizing many types of faults.