Protein topology, which refers to the arrangement of secondary structures of proteins, has been extensively investigated to examine its role in protein folding. However, recent studies show that topology alone cannot account for the variation of folding behaviors observed in some proteins of the same structural family. In a recent work, we showed that the native structure of the second beta hairpin of protein G predicts a folding mechanism that is different from topology-based models. Here, we continue to examine how much one can learn about folding mechanism from native structure. This work focuses on fragment B of Staphylococcal protein A (BpA) - a three-helix (H-1, H-2, and H-3) bundle protein. Using a recently developed all-atom (except nonpolar hydrogen) G (o) over bar model interacting with simple discontinuous potentials, the folding of the model BpA was observed in 112 out of 249 trajectories within 50 h of CPU times on a Pentium PC (1 GHz). The model successfully captured several specific properties of BpA that have been observed experimentally. These include the higher stability of H-3 compared to H-1 and H-2, and the higher stability of the H-2-H-3 microdomain compared to the H-1-H-2 microdomain. These specific details were not produced by a topology-based square-well model of BpA. Thus, the result further supports the important role of sidechain packing in determining the specific pathway of protein folding. Additional 96 000 short simulations were performed to locate the transition states of the two folding pathways. The limitation of the G (o) over bar model and its possible improvement are also discussed.