Problem-solving strategies are crucial in learning programming. Owing to their hidden nature, traditional methods such as interviews and questionnaires cannot reflect the details and differences of problem-solving strategies in programming. This study uses the Hidden Markov Model to detect and compare the problem-solving strategies of different groups in an interactive programming environment. The results suggest that high- and low-performance students have significant differences in their problem-solving strategies in programming. High-performance students had more "blank behaviors" in programming than low-performance students in video recordings. Low-performance students spent more time "searching teaching materials" than high-performance students. In the transfer task, high-performance students began the task by "identifying the problem," while low-performance students were involved in the "implementing of strategies." Additionally, high- and low-performance students improved from basic to transfer tasks. These findings shed light on why students performed differently in programming and how and when teachers needed to provide instructions to students in programming education.