The boundary element code MULSIM/NL was modified to accommodate a larger number of elements in the mesh, and a stress-versus-strain material model defined by a piecewise linear function, called the five-point material model, was added and verified. This updated version of MULSIM/NL is named MulsimNL/Large. MulsimNL/Large and FLAC 3D , the latter using an unverified and uncalibrated caving model, calculated gob loading that was in a reasonable range. However, better estimates of gob loading are necessary to proceed further with full model calibration. Detailed pillar models using FLAC 3D suggested that alternatives to the Mark-Bieniawski pillar strength model should be considered, especially for pillars with a width-to-height ratio greater than eight. A Holland-Gaddy equation fit to data from FLAC 3D models consisting of average pillar strength versus pillar width-to-height ratio showed better ability to describe model results than did the Mark-Bieniawski strength equation. Experience with the detailed pillar models also suggests that stratigraphy should be modeled, wherever possible. Borehole pressure cell (BPC) trends and comparison of model-calculated load transfer distance with equivalent load transfer distance determined from measurements of first arrival of mining-induced abutment stress strongly suggest that commonly used yield models were inadequate to describe seam material behavior near the ribs at Mine A. A MulsimNL/Large elastic model with the coal modulus reduced in the outer two rings to 60% and 80% of full value, respectively, provided the best match of "measured" versus model-calculated load transfer distance derived from stress increase at each BPC site. In considering appropriate strength models that might sufficiently simulate this behavior, measurements indicate that ribs are sometimes able to take much more stress than that calculated using most common yield strengths models, suggesting that strain-hardening might be an important part of the coal pillar behavior near pillar ribs. Observations in these two case studies (Mines A and B) suggest that geologic structure changes, not represented in the models, were important for predicting stress and instability of ground. Observations suggest that prior knowledge of structure change and individual structure properties would be necessary to adequately simulate its effect on stresses and failures with a particular mining layout.