Net ecosystem exchange (NEE) between terrestrial ecosystem and atmosphere can be directly observed by eddy covariance flux observation system, but we need to get accurately ecosystem gross primary productivity (GPP) and respiration (Re) under different time scales in order to get insights into carbon cycle process. This paper analyzed the eddy carbon flux and meteorology measurement data of a mid-subtropical planted coniferous forest at Qianyanzhou station from 2003 to 2009, and explored the impacts of two different NEE partition methods on estimation of ecosystem GPP and Re under different time scales. Results indicated that ecosystem Re and GPP estimated by different eddy flux partition methods showed similar seasonal dynamics, both of which reached the peaks in July or August of growing season. However, the annual Re and GPP estimated by nonlinear regression model were 2%-28.6% and 1.6%-23% higher than those estimated by light response curve model, respectively. The maximum annual ecosystem respiration difference between two methods existed in 2006 (317.6 gC·m^-2·a^-1), and the maximum monthly ecosystem respiration difference mostly appeared in August or September. Also, we found that environmental factors significantly affect differences between two derived Re (or GPP) with various time scales. For example, the vapor pressure deficit and photosynthetic active radiation were found to explain 63% and 60% of the ecosystem respiration difference between two methods at daily time scale, respectively. Moreover, precipitation, vapor pressure deficit and photosynthetic active radiation could explain 48%, 85% and 89% of the ecosystem respiration difference between two methods at monthly time scale. Third, 78% of the ecosystem respiration difference between two methods could be explained by the photosynthetic active radiation at yearly time scale. It means that the photosynthetic active radiation could explain the most of Re difference between two methods under three time scales. In spite of the wide application of the nonlinear regression model, it was necessary to allow for the light response curve model to partition the carbon flux of that month whose monthly PAR is about 905mol·m^-2·mon^-1 and the vapor pressure deficit is around 1.18 KPa as a reference, compared with those partitioned by the nonlinear regression model. Furthermore, the research got access to improvements on the partition results of ecosystem carbon flux and reduced the partition uncertainty.