The amplitude of a seismic refraction event is determined by the properties of rocks through which the seismic waves travel, the amplitude of the shot and the offset at which the refraction is recorded. A surface-consistent, non-linear inversion scheme, which uses the Levenberg-Marquardt algorithm, has been developed which aims to extract near-surface rock properties from the measured refraction amplitudes. Perhaps the most important challenge to extracting a unique, geologically plausible solution is in determining initial values of control parameters which dictate how the solution is allowed to progress. If these control parameters are not tailored specifically to the problem in question, convergence to a solution can be very slow and even fail to get off the ground in some cases. Comparison between results obtained using default control parameters, compared to those obtained using control parameters specifically tailored to the problem, shows a reduction in the error between the true observations and the model-generated observations, while retaining fidelity in the solution. A reformulation of the problem significantly reduces the error but fidelity is apparently compromised. Comments are made on how to achieve an optimal middle ground.