Detecting the presence of musical sounds in broadcast audio is important for content-based indexing and retrieval of auditory and visual information in radio and TV programs. In this paper, we propose a novel approach for musical sounds detection in broadcast audio based on the analysis of the characteristic feature of musical tones, pitch tuning. A spectral analysis method is presented for detecting the evidence of pitch tuning in the audio signal. Unlike the existing methods for discriminating speech and music, the proposed technique is not limited by inadequate training data, and it can deal with the case of music mixed with speech. In addition, the technique can be efficiently implemented for real-time application. Experiments based on TRECVID data set have shown good performance of the proposed technique.