Ecosystem carbon budgets depend on there being good representative surface flux observations for all land use types during the entire diurnal cycle. In calm conditions that often occur at night, especially in areas of small roughness (such as pastures), ecosystem respiration rate is poorly measured using the eddy covariance (EC) technique. Nocturnal vertical profiles of temperature, humidity and winds were observed using tethered balloon soundings in a pasture in the eastern Amazon during two campaigns in 2001. The site is characterized by very weak winds at night, so that there is insufficient turbulence for the EC technique to determine fluxes accurately. To compensate, the time evolution of the profiles is used to determine surface fluxes at early morning and these are compared with those observed by EC at a nearby micrometeorological tower. The nocturnal boundary layer thickness h is determined as the height to which the surface fluxes must converge so that energy budget closure is achieved. The estimated values range from 30 m, around 22:00 hours LST, to more than 100 m just before dawn. These are in good agreement with the observed thickness of a frequently observed fog layer during the middle of the night. During the early portion of the night, when the accumulation layer is shallow, there is appreciable decrease of dCO2/dt with height. On calm nights, CO2 accumulation rate is larger near the surface than at higher levels. On windier nights, this accumulation rate is vertically uniform. Hence, extrapolation of tower profiles for estimating fluxes must be done carefully. Although uncertainties remain large, an alternate approach to the EC method is described for measuring nighttime surface CO2 fluxes under stable atmospheric conditions.