The analytic energy gradient for the singles and doubles coupled cluster method including a perturbative correction due to triple excitations [CCSD(T)] is formulated and computatonally implemented. Encouraged by the recent success in reproducing the experimental equilibrium structure and vibrational frequencies of ozone, the new CCSD(T) gradient method is tested with two other "difficult" quantum chemistry problems: FOOF and Cr2. With the largest basis set employed in this work [triple zeta plus two sets of polarization functions (TZ2P-integral)] at the CCSD(T) level of theory, the predictions for the O-O and O-F bond lengths in FOOF are 1.218 and 1.589 angstrom, respectively. These figures are in good agreement with the experimental values 1.216 and 1.575 angstrom. Based on CCSD calculations with even larger basis sets, it is concluded that the error of 0.014 angstrom in the O-F bond length at the TZ2P-integral/CCSD(T) level of theory is due to the remaining basis set deficiency. On the other hand, the CCSD(T) prediction for the equilibrium bond length of Cr2 (1.604 angstrom), obtained with a large (10s8p3d2-integral-1g) basis set capable of achieving the Hartree-Fock limit, is still 0.075 angstrom shorter than experiment, clearly indicating the importance of higher than connected triple excitations in a single-reference treatment of this particular problem.