Abstract : Software error detection is about finding and prediction software logical error before deliver it to customers by some of automatic approaches. Some of logical errors are hidden in source code and cannot be found simply and no piece of software is free of logical errors. As software is written by humans, errors can always occur. So, finding and prediction of errors is one of most important issues in software development cycle. Program logical error localization and program testing are two of the most important sections in software engineering. Programmers or companies that produce programs will lose their credit and profit effectively if one of their programs delivered to a customer has any drawback. Nowadays there are many methods to test a program. Invariant are program valuable properties and relations which are true in all executions. This paper suggests a framework to localize the program logical errors by extraction of knowledge from invariants using a clustering technique. Keywords: Invariant Rules, Dynamic Detection, Variable Relations, Software engineering, Verification