The response of a concrete filled, steel pipe pile-to-concrete pile cap connection subjected to extreme lateral loads was experimentally and analytically investigated in this project. This connection is part of a bridge support system used by the Montana Department of Transportation that consists of a linear array of piles connected at the top by a concrete pile cap. Five 1/2 size models of this connection were tested to failure under monotonically increasing and/or cyclic lateral loads. The primary attribute of the connection that was varied between tests was the amount and layout of the reinforcing steel in the pile cap. The depth of embedment of the pipe pile in the cap was held constant. The first tests were done on lightly reinforced pile cap cross-sections, and failure occurred in the pile caps due to tensile cracking of the concrete and yielding of the reinforcing steel adjacent to the pile. In subsequent connections, the amount of reinforcing steel in the cap was increased, and its arrangement was modified, until a plastic hinge occurred in the pipe pile before failure of the cap occurred. The behavior of each connection was analyzed using hand calculations, strut and tie models, and solid finite element models. The hand calculations accurately predicted the nature of the failure mechanism for each connection, but only poorly predicted the magnitude of the failure load. The strut and tie models used in this investigation were created and analyzed using conventional structural analysis software. The resulting models offered significant detail relative the response throughout the pile cap, but were unable to fully represent yielding of the reinforcing steel and the attendant redistribution of stresses within the cap. Sufficiently promising results were obtained relative to predicting the load and location at which inelastic behavior will initiate, that this analysis methodology possibly should be pursued further. Finally, though finite element models were not successfully used to model the damage cycle through cyclic loads as originally hoped, they did prove useful for extracting 3D information leading up to a state of permanent damage. They also show immediate promise for modeling responses to monotonic load conditions, particularly for analysis where concrete damage is not the controlling failure mechanism.