We have previously demonstrated the discrimination of two layers of soft tissue, specifically normal breast tissue overlying breast tumor, using spatially offset Raman spectroscopy (SORS). In this report, a Monte Carlo code for evaluating SORS in soft tissues has been developed and compared to experimental results. The model was employed to investigate the effects of tissue and probe geometry on SORS measurements and therefore to develop the design strategies of applying SORS for breast tumor surgical margin evaluation. The model was used to predict SORS signals for different tissue geometries difficult to precisely control experimentally, such as varying normal and tumor layer sizes and the addition of a third layer. The results from the model suggest that, using source-detector separations of up to 3.75 mm, SORS can detect sub-millimeter-thick tumors under a 1 mm normal layer, and tumors at least 1 mm thick can be detected under a 2 mm normal layer.