Our group developed a force spectroscopy capability for use with our near-atomic level Upside MD package, and applied it to the vertical and lateral pulling of bacteriorhodopsin (bR) and GlpG, respectively. The efficiency of our model and software allows for several orders of magnitude slower pulling speeds, which are closer to experiment, than possible with traditional all-atom methods. For both proteins, unfolding is not cooperative, as we observe long-lived intermediates. To uncover the dominant factor(s) contributing to the rupture forces, we selectively tuned individual interactions. For vertical pulling of bR perpendicular to the membrane, hydrogen bond strength has the strongest effect, whereas other non-bonded protein and membrane interactions have only moderate influence, except for the last helix, where we predict a stronger influence of the membrane potential. The up-down topology of the transmembrane helices causes them to be pulled out as pairs, except for the 7th helix, which is pulled out by itself. The rate-limiting event is a critical loss of hydrogen bonds and the ejection of the first helix, which then propagates tension to the second helix. Surprisingly, the pulling of the charged linkers across the membrane has minimal influence. For lateral pulling of GlpG, the rate-limiting events instead correspond to the separation of the helices within the membrane, whereas hydrogen bonds are generally, but not always, lost afterward. This study emphasizes that the unfolding mode has a large effect on the region of the energy surface sampled by the protein.