Accurate estimation of ecosystem respiration (R eco ) in forest ecosystems is critical for validating terrestrial carbon models. Continuous eddy covariance measurements of R eco were conducted in a coniferous and broad-leaved mixed forest located in Dinghushan Nature Reserve of southern China. R eco was estimated and the controlling environmental factors were analyzed based on two years' data from 2003 to 2004. Major results included that: (1) R eco was affected by soil temperature, soil moisture, canopy air temperature and humidity, where soil temperature at 5 cm depth was the dominant factor. (2) The exponential equation, Van't Hoff equation, Arrhenius equation and Lyold-Talor equation can be used to describe the relationship between R eco and temperature factors with similar statistical significance, while Lyold-Talor equation was the most sensitive to the temperature index (Q 10 ). (3) The multiplicative model driven by soil temperature (T s ) and soil moisture (M s ) was more corresponsive to R eco , which explained that there were more R eco variations than Lyold-Talor equation, both for higher and lower M s . However, there was no statistical difference between the two models. (4) Annually accumulated R eco of the mixed forest in 2003 was estimated as 1100–1135.6 gC m −2 a −1 by using daytime data, which was 12%–25% higher than R eco (921–975 gC m −2 a −1 ) estimated by using nighttime data. The results suggested that using daytime data to estimate R eco can avoid the common underestimation problem caused by using eddy covariance methods. The study provides a basic method for further study on accurate estimation of net ecosystem CO 2 exchange (NEE) in the coniferous and broad-leaved mixed forest in southern China.