Which one is better for geometry optimization, B3LYP/4-31G or HF/6-31G*?

 I have been doing calculations on geometry optimizing of some organic
 molecules with great ring constraints. I tried some semi-empirical
 methods, for example, MNDO and PM3, but the results were rather
 different with the corresponding XRD crystal datas. So I have to use
 non-empirical ab initio methods to do the same work.
 My question is which method is better, B3LYP/4-31G or HF/6-31G*?
 With best wishes
 Y. Fan
 Yubo Fan                         Email: yubofan : at : guomai.sh.cn
 Organic Synthesis Lab                   yubofan : at : fudan.edu.cn
 The Department of Chemistry
 Fudan University                 Phone: 8621-65648139
 No. 220 Handan Road              Fax:   8621-65641740
 Shanghai, 200433
 P. R. China