A large set of tower data was used to identify the gap that separates small-scale turbulence and mesoscale structures in the cospectra of surface fluxes. The cospectra were obtained using a multi-resolution decomposition algorithm. The gap time scale tau (g) was found by fitting a fifth-order polynomial to the cospectra and identifying special points occurring after the peak at small scales. In unstable conditions (day) tau (g) was found to fall as the mean wind speed increased, while no such dependence was observed in stable conditions (night). The gap scale was found to change very weakly with stability both in moderately stable and moderately unstable conditions, with a sharp drop from about 1100 to 250 s occurring in near-neutral conditions. The vertical fluxes computed at different averaging intervals were found to correlate exceptionally well with each other, the scatter being somewhat larger during the night. Although considerable discrepancy may occur for individual records, when averaged over 10 months, the difference in the flux estimated at 7 to 109 min intervals never exceeded 4%, which is comparable or less than the instrumental error..