The energy of a signal decreases with increasing propagation distance. When two sensors at different locations are used to collect the signals, each signal has its own energy attenuation ratio. In this paper, the energy attenuation feature of signals was employed to estimate the number of sources. For sources located in different frequency bands, the linear time-frequency transform method can be utilized to compute the energy corresponding to different frequency bands. Due to different energy attenuation ratios for each source, several cluster features will appear on the 2-D plot. Then, energy peak detection can be used to determine the number of sources. The proposed method can easily implement the experimental program and algorithm, and can be used to estimate an arbitrary number of sources. The effectiveness was verified by theoretical analysis, simulation data and experimental data, such as gearbox vibration signals, sound pressure signals and modal signals.