The analysis of a song structure is the first step for any bioacoustic and communication study in acoustic animals. However, the structure of the advertisement song is often arbitrarily defined for description of song characters in many cases. We developed a new method of identifying song structures, which relied on the strucchange package in R-program, a robust test capable in detecting changes of structures with unknown timing and complexity. Time series of peak frequency and peak amplitude are obtained from a recorded song. Then, a generalized linear regression model based on those two time series would be tested for consistency of regression coefficients. The package identifies locations in a song, when there are significant structural changes in the sequence. We applied this method in analyzing advertisement songs of three cicada species in Korean peninsula: Cryptotympana atrata, Hyalessa fuscata, and Meimuna opalifera. Songs of C. atrata were divided into two parts with no repeated structures. The calling songs of H. fuscata were divided into four main parts, in which two parts were repeated alternatively and the other two were produced once at the end of the song. Furthermore, each part was further divided into echemes of distinct low and high frequencies. The songs of M. opalifera were characterized by five parts; each part was produced once and consisted of multiple frequency- and amplitude-modulating echemes. Our method provides an ease and effective way in analyzing acoustic characteristics of animal’s songs, especially those with complex structures.