Several workers have developed cross section balancing schemes for extensional structures. Using these geometric models, structural interpreters can predict normal-fault geometries at depth by specifying the shape of a folded bed in the hanging wall of a normal fault, the deformation mechanism that accommodates the folding, the footwall counterpart of the folded hanging-wall bed, and the dip and location of the fault segment between the hanging-wall and footwall beds. We have systematically determined how uncertainties in these input parameters affect the prediction of normal-fault shape and detachment depth. Our sensitivity analysis shows that reasonable uncertainties in the fold shape near the normal fault, the deformation mechanism, the footwall correlation, and the fa lt dip produce large uncertainties in the predicted fault geometry. Reasonable uncertainties in the fold shape far from the normal fault and the fault location, however, produce only small uncertainties in the predicted fault geometry. Our work suggests that if seismic and well data are lacking or of poor quality, then geometric models cannot provide unique answers to questions about the geometry of a normal fault at depth. Geometric models, however, can provide a range of possible fault geometries.