We can now access to large music databases. However, it is not easy to find favorite music pieces that the user does not know yet from those databases. Therefore, the demand for the music recommendation system is increasing. The purpose of this study is to construct a system to recommend music pieces that suit each user's preferences from their unknown music pieces. In the proposed method of this study, first, we create a melody for each user using Interactive Genetic Algorithm (IGA). Next, we recommend music piece that contains parts similar to the created melody using the music recommendation system. In the recommendation, relative key is used. To confirm the effectiveness of the proposed method, we conducted 2 listening experiments with 10 subjects. In the melody creation experiment, the average subjective fitness gradually increased, and significant differences between the first generation and some generations including the final generation was observed. On the other hand, in the music recommendation experiment, effectiveness could not be confirmed. The failure of the recommendation might be caused from shortage of the size of the music database: it contains only 100 pieces. Therefore, the music piece which was sufficiently similar to the created melody could not be recommended. From the above results, expansion of the music database is discussed as a future task.