An extremely careful inquiry is made into the possibility of predicting bond lengths in condensed aromatic hydrocarbon molecules. Agreement with the best experimental values, such as those of Robertson, Abrahams, White, Mathieson and Sinclair, is fairly easily obtained to an accuracy of about 0$\cdot $02 angstrom. This shows that the concept of fractional bond order may quite properly be used to infer bond lengths. Both the molecular-orbital and resonance methods are equally good for this purpose. Predictions to within less than 0$\cdot $02 angstrom require the introduction of new factors usually neglected. No less than five such factors are discussed: (a) electrostatic forces, arising from possible differences in electronegativity of the various carbon atoms, (b) changes of bond orders due to electronegativity differences, (c) variation of resonance integrals with bond length, (d) obtaining a self-consistent set of resonance integrals, (e) inclusion of configuration interaction. Correction terms which result from these improvements lie between 0 and 0$\cdot $015 angstrom, and are not all of the same sign. It is unlikely therefore that this type of analysis will be able to give confident predictions of bond lengths to less than 0$\cdot $01 angstrom.