Abstract Fog deposition, precipitation, throughfall and stemflow were measured in a windward tropical montane cloud forest near Monteverde, Costa Rica, for a 65‐day period during the dry season of 2003. Net fog deposition was measured directly using the eddy covariance (EC) method and it amounted to 1·2 ± 0·1 mm day −1 (mean ± standard error). Fog water deposition was 5–9% of incident rainfall for the entire period, which is at the low end of previously reported values. Stable isotope concentrations (δ 18 O and δ 2 H) were determined in a large number of samples of each water component. Mass balance‐based estimates of fog deposition were 1·0 ± 0·3 and 5·0 ± 2·7 mm day −1 (mean ± SE) when δ 18 O and δ 2 H were used as tracer, respectively. Comparisons between direct fog deposition measurements and the results of the mass balance model using δ 18 O as a tracer indicated that the latter might be a good tool to estimate fog deposition in the absence of direct measurement under many (but not all) conditions. At 506 mm, measured water inputs over the 65 days (fog plus rain) fell short by 46 mm compared to the canopy output of 552 mm (throughfall, stemflow and interception evaporation). This discrepancy is attributed to the underestimation of rainfall during conditions of high wind. Copyright © 2010 John Wiley & Sons, Ltd.