Anecdotal evidence shows that in computer programming courses plagiarism is a widespread problem. With the growing number of students in such courses, manual plagiarism detection is impractical. This requires instructors to use one of the many available plagiarism detection tools. Prior to choosing one of such tools, a metric that assures the chosen plagiarism detection tool is effective for novice programmer plagiarism is required. It must be able to detect the difference between accidental similarities and plagiarized code. In this research we looked for such metrics by an experimental approach that was carried out in three phases. In the first phase we learned the possible ways to plagiarize, which was done by interviewing students. Categorizing the known methods lead us to the most popular plagiarism approach that was applied by the students. Therefore an experiment was designed to simulate the real plagiarism situation. Data that was gathered in this phase were evaluated against real plagiarism data to ensure the results. The result showed that object-orientation concepts were of little interest for students to plagiarize; instead converting basic structures and syntax were more popular. This means that a source code which is not original is less likely to be different in terms of inheritance structure, encapsulation, or polymorphism from its original source code. This implies that an effective plagiarism detection system that is used for undergraduate teaching purposes should be able to identify similar simple structures rather than focusing on some more advanced options of languages such as inheritance. It should be noted that the conducted experiments included both freshmen and senior students and the programming language that was used for the experiment was Java, which is an object-oriented language. Therefore the results can be generalized to all computer science undergraduates who program using an object oriented language. (Contains 2 tables.)