Waveform tomography is commonly tested using numerically generated synthetic seismic data, before the method is applied to field seismic data. However, there are often noticeable differences between idealized synthetic data and real field data, and many factors in the field data, such as noise, irregular source/receiver geometry, affect the inversion solutions. For exploring the potential of reflection seismic waveform tomography, we presented a more realistic test than the synthetic data test, by applying it to physical modelling data, to reconstruct a laboratorial model with complex velocity variation. First, we provided a formulation of the perfectly matched layer absorbing boundary condition, associated with the second-order acoustic wave equation, in order to suppress artificial reflections from subsurface model boundaries in seismic waveform simulation and tomography. Then, we demonstrated the successful implementation of a layer-striping inversion scheme applicable to reflection seismic waveform tomography. Finally, we confirmed the effectiveness of frequency grouping, rather than a single frequency at each iteration, a strategy specifically for the frequency-domain waveform tomography.