Seismic refraction tomography is an alternative to conventional seismic refraction analysis methods. While the limitations and potential pitfalls of conventional refraction methods are well-known the same is not true for refraction tomography. As refraction tomography becomes more widely used, the need to know and understand its capabilities as well as its limitations becomes more critical. In this study we created eight representative models for use in evaluating three commercially available codes as well as refraction tomography in general. These models range from simple two-layer or dipping-layer problems to more complicated models designed to represent features of karst terrains. We demonstrate quantitatively and qualitatively that all three codes perform at a similar level, although each has strengths and weaknesses. Refraction tomography performs well in many situations where conventional methods fail, e.g., where lateral or vertical gradients compose a significant component of the velocity structure.