A research on the temporal preference in course-of-action planning is presented, focusing on the scheme to implement planning with inconsistent temporal preference. The difficulties and innovations lie in two aspects. Firstly, considering the representation issue, a unified binary format is established for temporal preference and temporal constraints. Secondly, argumentation is utilized to develop a deduction framework in order to sort out the conflicts inside the constraint/preference set, achieving a subset of constraints/preferences with maximum consistence. A plan generation strategy was then proposed to acquire segments of plan via mapping, and to combine the segments into a complete plan. A case study was done in detail to show the effectiveness of the scheme.