Abstract: Test case generation is one of the most important and costly steps in software testing. the techniques for automatic generation of test cases try to efficiently find a small set of cases that allow an adequacy criterion to be fulfilled, thus, reducing the cost of software testing and resulting in more efficient testing of software products. In this study, we analyze the application of different machine learning methods for automatic test case generation task. We describe in this study how these algorithms can help in white-box testing. Different algorithms, consists of random, GA, MA and a proposed hybrid method called GA-NN, are then considered and studied in order to understand when and why a learning algorithm is effective for a testing problem. For the experiments we use a benchmark program, called Triangle classifier. Finally, the evaluations and some open research questions are given.