In this paper, a promising technique for milk analysis based on an ultrasound spectroscopy is presented and tested. Particularly, it employ a wide-band random multi-sine as excitation signal to analyse the system responses aiming to obtain its transfer function. This signal assure a better utilization of the signal spectrum, compacting the energy in the frequency band of interest. An experimental setup has been developed to perform polydisperse system analysis with different ultrasonic techniques. Six different milk types were analysed, allowing to evaluate different fat, carbohydrate and protein concentrations. Results showed the transfer functions amplitude and phase variations for each type of milk studied.