Nonlinear finite element (FE) collapse pressure predictions are compared to experimental results for submarine pressure hull test specimens with and without artificial corrosion and tested to collapse under external hydrostatic pressure. The accuracy of FE models, and their sensitivity to modeling and solution procedures, are investigated by comparing FE simulations of the experiments using two different model generators and three solvers. The standard FE methodology includes the use of quadrilateral shell elements, nonlinear mapping of measured geometric imperfections, and quasi-static incremental analyses including nonlinear material and geometry. The FE models are found to be accurate to approximately 11%, with 95% confidence, regardless of the model generator and solver that is used. Collapse pressure predictions for identical FE models obtained using each of the three solvers agree within 2.8%, indicating that the choice of FE solver does not significantly affect the predicted collapse pressure. The FE predictions are found to be more accurate for corroded than for undamaged models, and neglecting the shell eccentricity that arises due to one-sided shell thinning is found to significantly decrease the resulting accuracy of the FE model.@en