How To Find Molecular Geometry

Understanding the Basics

Finding molecular geometry comes down to three things: a correct Lewis structure, counting electron domains around the central atom, and matching that count to a shape. The underlying model is VSEPR — valence shell electron pair repulsion. The name is slightly misleading because it implies the geometry is driven by "pairs" only, when in practice you're really counting regions of electron density regardless of whether they're bonding pairs, lone pairs, or part of a multiple bond. Draw the Lewis structure first. Get the total valence electron count right, place the least electronegative atom in the center, connect the surrounding atoms, and then distribute remaining electrons to satisfy octets. If you run out of electrons before the central atom has an octet, form double or triple bonds. This step is where most mistakes happen, and if the Lewis structure is wrong, every geometry afterward is wrong too. Next, count electron domains. A single bond counts as one domain. A double bond counts as one domain. A triple bond counts as one domain. A lone pair counts as one domain. That's it. You do not add extra domains for the second or third bond in a multiple bond. A carbonyl group is one domain, not two.

Then match the domain count to the electron geometry. Two domains gives linear. Three gives trigonal planar. Four gives tetrahedral. Five gives trigonal bipyramidal. Six gives octahedral. The molecular geometry is what you get when you ignore the lone pairs and look only at the positions of the atoms. Lone pairs occupy more space than bonding pairs, which means bond angles will often be smaller than the ideal values listed in textbooks. Water is the classic example — the electron geometry is tetrahedral, but the molecular geometry is bent with an actual bond angle of about 104.5 degrees, not the 109.5 you'd expect from a perfect tetrahedron. Here is a table you can keep open while you work. It covers the most common cases. 2 bonding domains, 0 lone pairs: linear, 180 degrees.
3 bonding domains, 0 lone pairs: trigonal planar, 120 degrees.
3 bonding domains, 1 lone pair: bent, slightly less than 120 degrees.
4 bonding domains, 0 lone pairs: tetrahedral, 109.5 degrees.
4 bonding domains, 1 lone pair: trigonal pyramidal, slightly less than 109.5 degrees.
4 bonding domains, 2 lone pairs: bent, about 104.5 degrees.
5 bonding domains, 0 lone pairs: trigonal bipyramidal, 90 and 120 degrees.
5 bonding domains, 1 lone pair: seesaw, angles compressed from ideal.
5 bonding domains, 2 lone pairs: T-shaped.
5 bonding domains, 3 lone pairs: linear.
6 bonding domains, 0 lone pairs: octahedral, 90 degrees.
6 bonding domains, 1 lone pair: square pyramidal.
6 bonding domains, 2 lone pairs: square planar.

Tools and Software

If you are doing this by hand for homework, the method above is fine. If you are working with anything larger than about five atoms, manual drawing becomes unreliable and slow. Several free tools can generate 3D structures and report geometries automatically. MolView (molview.org) is the quickest option. You paste a SMILES string or InChI key, and it builds the molecule with a 3D viewer. It does not always place lone pairs visibly, but the atom positions and approximate bond angles are there. For a quick check on whether your predicted geometry matches computational output, this is sufficient. Avogadro (avogadro.cc) is more capable. It includes geometry optimization using semi-empirical methods. You type in the composition or build the structure manually, run a quick optimization, and the software adjusts bond lengths and angles to a local energy minimum. The default level of theory is fast but not highly accurate. It will give you a reasonable geometry in seconds for a small molecule. For production-quality results you would switch to DFT, but that takes longer and requires more setup.

PubChem (pubchem.ncbi.nlm.nih.gov) is not a builder but a database. You can search for a compound and find its 2D and 3D conformers along with experimental and computed data. If you need a reference geometry rather than generating one yourself, this is often the most reliable source because many entries include X-ray crystallography data. WebMO offers a browser-based interface to quantum chemistry programs. It is more involved than MolView but gives you control over the method and basis set. If you need geometries for a publication or thesis, this is closer to what you would actually use in a research setting. The learning curve is steeper, and setting up a calculation correctly requires understanding which functional and basis set combination is appropriate for your system.

A Problem I Ran Into and How I Fixed It

I was checking the geometry of chlorate, ClO3, for a problem set a few years ago. The Lewis structure gives chlorine three bonding domains and one lone pair, which according to the table should be trigonal pyramidal. I built it in MolView, ran a quick optimization in Avogadro with the MMFF94 force field, and the resulting structure looked distorted in a way that did not match a clean trigonal pyramid. The O-Cl-O angles were not consistent, and the chlorine atom sat noticeably off the expected centroid. The issue was not with VSEPR itself. Chlorate has significant resonance across three equivalent Cl-O bonds, and the single force field used in the initial optimization did not handle the partial double-bond character well enough. I switched the optimization to the UFF force field, which performed better for inorganic oxyanions, and the geometry stabilized into something much closer to the expected trigonal pyramidal shape with angles around 106 to 107 degrees. Later I ran a semi-empirical PM6 calculation and got consistent results. The takeaway is that force field choice matters more than people usually admit, especially for molecules where resonance or d-orbital participation is a factor.

Counter-Intuitive Details Beginners Miss

First, molecular geometry and electron geometry are different things, and mixing them up is extremely common. Electron geometry includes lone pairs. Molecular geometry does not. The difference only matters when there is at least one lone pair on the central atom, but when it matters, it matters a lot. Saying water is tetrahedral is technically wrong if you are being asked for molecular geometry. It is bent. The electron geometry is tetrahedral. Second, multiple bonds do not create additional domains. A carbon in a carboxylate group has three domains — two single bonds to oxygen (one is technically a resonance hybrid, but the domain count is still three) and one bond to the rest of the molecule — giving trigonal planar geometry around that carbon. Students sometimes count each bond in a double bond separately and end up with the wrong shape. Third, VSEPR breaks down for certain classes of compounds. Transition metal complexes are the biggest example. Crystal field theory and ligand field theory are needed there because d-orbital splitting dominates the geometry, not simple electron domain repulsion. Square planar complexes like PtCl4^2 cannot be predicted reliably with VSEPR alone. Hypervalent molecules like SF6 also push the model, though VSEPR handles them acceptably with six domains and octahedral geometry. The real failure point is when you have competing electronic effects, such as in molecules with strong Jahn-Teller distortions, where the predicted symmetric geometry is never observed.

Limitations You Should Know About

VSEPR is a heuristic, not a law. It gives you a good first approximation for main-group compounds with a single central atom and no unusual electronic effects. It fails or becomes unreliable in several scenarios: Systems with multiple central atoms connected to each other require you to treat each center independently and then combine the results. The overall shape is not a single geometry but a collection of local geometries. This is manageable but easy to mess up if you pick the wrong central atom. Radicals are problematic. An unpaired electron counts as a domain in some treatments but not others, and the literature is inconsistent. NF2 radical is one case where the geometry is debated because the single electron behaves differently from a full lone pair.

Large or flexible molecules do not have a single fixed geometry. They have conformers, and the lowest-energy conformer depends on the environment. A gas-phase geometry from computation can differ from a solid-state geometry from X-ray diffraction because packing forces distort the structure. If you need accurate geometries for any of the above cases, or for anything involving transition metals, VSEPR is not the right tool. You should use quantum chemistry software instead. ORCA, Gaussian, and GAMESS are standard options. ORCA is free for academic use and relatively straightforward to set up. A typical DFT geometry optimization with B3LYP and a def2-SVP basis set on a moderate-sized organic molecule takes under an hour on a modern laptop. For a transition metal complex, expect longer and a higher chance of convergence issues that require manual intervention.

Quick Reference for Common Molecules

CO2: two domains, linear.
BF3: three domains, trigonal planar.
CH4: four domains, tetrahedral.
NH3: four domains, trigonal pyramidal.
H2O: four domains, bent.
PCl5: five domains, trigonal bipyramidal.
SF4: five domains, seesaw.
XeF4: six domains, square planar.
SO4^2: four domains, tetrahedral. Keep this list nearby while you practice. The patterns repeat, and after you have worked through enough examples, you will stop needing it. The main thing to watch for is getting the Lewis structure right in the first place. Everything else follows from that.