To clarify factors that affect efficiency of debugging, we need to investigate programers' intentions which continuously vary during. For the purpose, we have developed a system for observing programmers behavior by using an eye tracking device and audio and visual devices. By using the system, we have performed an experiment of observing bug identification processes. This paper presents results of the experiment,including the fact that collecting and analyzing multimodal data, such as motion of gazing point and record of verbal protocol, is effective in investigating programmers' intentions. It also includes several differences among observed processes which are likely to relate to the efficiency of bug identification.