Stress testing is a type of performance testing designed to determine the performance of system and to predict system's behavior under stressful loads. In this paper, we focus on this type of performance testing. We create a realistic performance test plan by identifying the key scenario for the application, the navigation paths for each scenario, the unique data for navigation paths, and user distribution along web applications scenario. In our proposed test, we determine the total number of users, application navigation paths, and then we distribute users along these navigation paths. We take in our account users think time, and data input and output. Once test begins, the users begin to send their requests to the web application the determined navigation paths; also it increases steeply as we planned in our proposed test plan. During that, the performance critical metrics like response time, throughputs, error rates and resource utilizations are automatically recorded then analyzed.  Our experimental results show that our proposed algorithm is robust for detections of errors location, percentages and reasons. It also detects the bottlenecks if exit. The experimental results also show with comparison the advantages of user load distribution on reducing the error percentage, the response time and the resource utilization.