CONSIDERING THE IMPORTANCE OF SOFTWARE SYSTEMS IN HUMAN LIFE, THEIR QUALITY ASSURANCE IS VERY IMPORTANT. FAULT LOCALIZATION IS ONE OF THE SOFTWARE TESTING STEPS, IT TRIES TO FIND THE EXACT LOCATION OF FAULT IN CODE. MOST OF AUTOMATIC FAULT LOCALIZATION TECHNIQUES USE COVERAGE INFORMATION AND RESULTS OF TEST CASES TO CALCULATE THE PROGRAM ENTITIES SUSPICIOUSNESS BY SIMILARITY COEFFICIENTS. THE SIMILARITY COEFFICIENTS DESIGNED BASED ON THE INSIGHT AND UNDERSTANDING OF DEVELOPERS FROM SOFTWARE SYSTEM AND THEY DO NOT HAVE THE SAME PERFORMANCE IN DIFFERENT SCENARIOS. TO OVERCOME WITH THIS PROBLEM, WE USE THE BACK PROPAGATION NEURAL network AND INVESTIGATE THE EFFECT OF weighted THE NEURAL network TO ACCURACY OF LOCATING FAULTS IN SOFTWARE PROGRAMS, BECAUSE THE BACK PROPAGATION NEURAL network IS SENSITIVE TO WEIGHT AND BY THE INITIAL PROPER WEIGHTS TO THE INPUT LAYER NEURONS CONNECTIONS, THE SEARCH SPACE TO ACHIEVE OPTIMAL WEIGHT IS DECREASING AND network ACCURACY IMPROVES. WE ANALYZE THE EFFECTIVENESS OF THE PROPOSED METHOD WITH RANDOMLY WEIGHTING THE INPUT LAYER NEURONS AND SOME BASIC AND EFFICIENT SIMILARITY COEFFICIENTS ON SIEMENS SUITE BENCHMARK. THE RESULTS SHOW THAT PROPOSED METHOD HAS A SATISFACTORY PERFORMANCE FOR THE SOFTWARE FAULT LOCALIZATION PROCESS.