Purpose To investigate an implementation method and the results of an inverse dose optimization algorithm, Gradient Based Planning Optimization (GBPO), for three-dimensional brachytherapy. Methods The GBPO used a quadratic objective function, and a dwell time modulation item was added to the objective function to restrict the dwell time variance. We retrospectively studied 4 cervical cancer patients using different applicators and 15 cervical cancer patients using the Fletcher applicator. We assessed the plan quality of GBPO by isodose lines for the patients using different applicators. For the 15 patients using the Fletcher applicator, we utilized dose-volume histogram (DVH) parameters of HR-CTV (D-100%, V-150%) and organs at risk (OARs) (D-0.1cc, D-1cc, D-2cc) to evaluate the difference between the GBPO plans and the IPSA (Inverse Planning Simulated Annealing) plans, as well as the GBPO plans and the Graphic plans. Results For the 4 patients using different applicators, the dose distributions are conformable. For the 15 patients using the Fletcher applicator, when the dwell time modulation factor (DTMF) is less than 20, the dwell time deviation reduces quickly; however, after the DTMF increased to 100, the dwell time deviation has no remarkable change. The difference in dosimetric parameters between the GBPO plans and the IPSA plans is not statistically significant (P0.05). The GBPO plans have a higher D-100% (3.57 +/- 0.36, 3.38 +/- 0.34; P0.01) and a lower V-150% (55.73 +/- 4.06, 57.75 +/- 3.79; P0.01) than those of the Graphic plans. The differences in other DVH parameters are negligible between the GBPO plans and the Graphic plans. Conclusions The GBPO plans have a comparable quality as the IPSA plans and the Graphic plans for the studied cervical cancer cases. The GBPO algorithm could be integrated into a three-dimensional brachytherapy treatment planning system after studying more sites.