Summary The scarcity of empirical data on leaf respiration ( R ) and its temperature sensitivity (e.g. Q 10 , defined as the proportional increase in R per 10 °C warming) causes uncertainty in current estimates of net primary productivity of tropical forests. We measured temperature response curves of R on 123 upper‐canopy leaves of 28 species of trees and lianas from a tropical forest in Panama and analysed variations in R and Q 10 in relation to other leaf functional traits. Respiration rates per leaf area at 25 °C ( R A ) varied widely among species and were significantly higher in trees than in lianas. R A was best predicted by a multiple regression model containing leaf phosphorus concentration, photosynthetic capacity and leaf mass per area ( r 2 = 0·64). The mean Q 10 value (2·4) was significantly higher than the commonly assumed value of 2·0. Q 10 was best predicted by the combination of leaf carbohydrate concentration and growth form (trees vs lianas) ( r 2 = 0·26). The night‐time leaf respiratory carbon flux from this tropical forest was calculated from these multiple regression models to be 4·5 Mg C ha −1 year −1 , with an estimated additional 2·9 Mg C ha −1 year −1 being released by respiration during the day. Trait‐based modelling has potential for estimating R , thus facilitating carbon flux estimation in species‐rich tropical forests. However, in contrast to global analyses, leaf phosphorus content was the most important correlate of R and not leaf nitrogen, so calibration of trait models to the tropics will be important. Leaf traits are poor predictors of Q 10 values, and more empirical data on the temperature sensitivity of respiration are critically needed to further improve our ability to scale temperature‐dependent respiration in species‐rich tropical forests.