The mechanisms that underlie the spontaneous and faithful assembly of virus particles are guiding the design of self-assembling protein-based nanostructures for biomedical or nanotechnological uses. In this study, the human immunodeficiency virus (HIV-1) capsid was used as a model to investigate what molecular feature(s) may determine whether a protein nanoparticle with the intended architecture, instead of an aberrant particle, will be self-assembled . Attempts of using the HIV-1 capsid protein CA for achieving the self-assembly of cone-shaped nanoparticles that contain CA hexamers and pentamers, similar to authentic viral capsids, had typically yielded hexamer-only tubular particles. We hypothesized that a reduction in the stability of a transient major assembly intermediate, a trimer of CA dimers (ToD), will increase the propensity of CA to assemble into cone-shaped particles instead of tubes. Certain amino acid substitutions at CA-CA interfaces strongly favored the assembly of cone-shaped nanoparticles that resembled authentic HIV-1 capsids. All-atom MD simulations indicated that ToDs formed by CA mutants with increased propensity for assembly into cone-shaped particles are destabilized relative to ToDs formed by wt CA or by another mutant that assembles into tubes. The results also indicated that ToD destabilization is mediated by conformational distortion of different CA-CA interfaces, which removes some interprotein interactions within the ToD. A model is proposed to rationalize the linkage between reduced ToD stability and increased propensity for the formation of CA pentamers during particle growth , favoring the assembly of cone-shaped HIV-1 capsid-like nanoparticles.