When a music audio to be analyzed is to be divided into a plurality of possible pattern segments based on estimated beat positions, it is divided in a plurality of different ways according to individual division models obtained by sequentially shifting the leading or first beat position of the possible pattern segments. Such division can provide plural sets of possible pattern segments with each of the sets corresponding to one of the division models. For each of the sets, comparison is made between individual possible pattern segments and individual reference performance patterns. For each of the possible pattern segments of the set, at least one reference performance pattern that matches the possible pattern segment is determined. Then, at least one combination of reference performance patterns is identified for each of the sets of possible pattern segments. One optimal combination is selected from among the identified combinations.