Rhythm is one of the main properties of Western tonal music. Existing content-based retrieval systems generally deal with melody or style. A few existing ones based on meter or rhythm characteristics have been recently proposed but they require a precise analysis, or they rely on a low-level descriptor. In this paper, we propose a mid-level descriptor: the Meter Class Profile (MCP). The MCP is centered on the tempo and represents the strength of beat multiples, including the measure rate, and the beat subdivisions. The MCP coefficients are estimated by means of the autocorrelation and the Fourier transform of the onset detection curve. Experiments on synthetic and real databases are presented, and the results demonstrate the efficacy of the MCP descriptor in clustering and retrieval of songs according to their metric properties.