電腦音樂作曲一直是電腦音樂研究者的夢想。本論文中，我們要探討的是從給定的多首音樂中，利用音樂作曲規則探勘的方式，找出給定音樂中的樣式，以產生具備這些音樂風格的新音樂。我們針對，音樂結構、和弦風格和音樂動機三項音樂特性做分析與探勘。在音樂結構部分，我們利用資料探勘技術分析音樂結構，並且學習音樂結構的特性。在旋律風格部分，我們分析每首音樂旋律的和弦以做為旋律的特徵，並從中探勘音樂的旋律風格。音樂動機部份，我們探勘出音樂動機，並探勘出音樂中音樂動機的重要性，以建立音樂動機挑選模型。最後根據音樂結構、和弦、音樂動機，三項音樂特性的學習結果產生整首音樂。在效果評估方面，我們採用類似Turing Test的方式，以測試機器產生的音樂和人所作曲的音樂之辨別率。結果顯示產生的音樂和人所作曲的音樂不易分辨。另外，實驗也顯示系統所產生的音樂在旋律及和弦上接近給定的音樂風格。 / Computer music composition has been the dream of the computer music researcher. In this thesis, we investigate the approach to discover the rules of music composition from given music objects, and automatically generate a new music object style similar to the given music objects. To discover the rules of music composition, the music is analyzed by addressing three music properties, music structure, melody style and motif. We exploit the data mining techniques to analyze music structure. For the melody style, chord is utilized to represent the feature of melody and melody style is discovered from chords of music objects. For the motif, modified repeating pattern finding algorithm is employed to discover the motives. Then, the motif selection model is constructed. A new music object is generated based on the discovered rules in terms of three music properties. To measure the effectiveness of proposed computer music composition approach, we adopt the method similar to the Turing test to test the discrimination between machine-generated and human-composed music. The result showed that it is hard to discriminate. Another experiment showed that the style of generated music is similar to the given music objects.