Weiner,Kollman,Nguyen,Case, J.Comput.Chem.,Vol.7,1986,230. This concerns following combinations: CC-CV-NB CC-CW-NA CC-NA-CR CC-NB-CR CR-NA-CW CR-NB-CV CT-CC-CV CT-CC-CW CT-CC-NA CT-CC-NB CV-CC-NA CW-CC-NA CW-CC-NB NA-CR-NA NA-CR-NB CW-NA-H CC-NA-HIn parm91 and parm94 the standard bonding angle is set to 117 or 120 degrees (which would be correct for six membered rings). This results in, to my opinion, too high energies for bond bending. The values taken form Weiner et al. seem to be more evident ranging from 105.3 to 109.9 degrees (which comes close to the angle in a regular pentangle)
There is a reason to make theta(0) for these sp2 atoms either 120 (for X-CA-X) or 117 (for X-NB-X): when we did normal mode analysis in JCC 1986, it was too easy to distort the rings from planarity using theta(0) values in the range of 105 degrees or so. By making theta(0) near 120, you make the energy cost for making these rings non-planar appropriately larger...also, it makes sense that the intrinsic electronic structure preference makes X-CA-X 120 and that five membered rings are strained in this regard, as they indeed are. If you made the theta(0) 105, they would be treated as "strainless", which is not correct. So there is method to our madness on this point.