The applicability of low-frequency sonar (2-16 kHz) to buried mine detection has been investigated. Experiments were performed on sound penetration into sediment, buried target detection and broadband multiple aspect classification. The results of the experiments are given in this report and compared with modelled results. One of the main results is our success in understanding the physical mechanisms contributing to subcritical penetration into sediment. It has been demonstrated that the evanescent wave was dominant in the lower frequencies 2-5 KHz of our bandwidth of interest 2-16 KHz. Roughness scattering dominates at higher frequencies (above 5 kHz) for our bottom type (RMS roughness 1.5 cm, cross-ripple correlation length 25 cm). Although roughness scattering has been shown to be one mechanism for explaining "anomalous" penetration into sediment, its potential for detection and classification of buried objects is unclear due to the low level and the lack of coherence of the received signals. It is demonstrated that sound speed variation with frequency, and therefore variation of the critical angle versus frequency, could exist for permeable sandy bottoms, which could influence the design of a buried mine sonar. The detection of buried targets is shown to be very effective at above the critical angle. At subcritical angles, detection becomes difficult. Significant gains in signal-to-reverberation ratio below critical angle were obtained either by emphasizing a relatively narrow band of frequencies at the lower end of the transmitted bandwidth (below 3 kHz) or by using a larger physical array or synthetic array processing which improve the sonar resolution. Simulations have shown that lower frequencies (of the order of 0.5-1 kHz) are essential to the detection of buried targets at low grazing angles and that the detection at those frequencies will only be effective for shallow buried targets.