Net Primary Productivity (NPP) is a key driver of ecosystem C balance. Its seasonal and annual variations can be measured directly at the stand level. However, estimating NPP on larger areas would require indirect methods such as: (i) process models, e.g. e-models based on the fraction of intercepted PAR (fIPAR) and on the light use efficiency (LUE = NPP/IPAR), or else models based on the water-use-efficiency (WUE = NPP/E, where E = evapo-transpiration); (ii) remote sensing, to estimate fIPAR (from the Normalized Difference Vegetation Index: NDVI) or else E (from the energy balance closure). However, two main impediments may interfere with such estimations of NPP: first, LUE and/or WUE may vary in time, and second, remote sensing may be unable to distinguish between the layers of the stands, which sounds critical for agroforestry systems. In a 20-year-old coconut grove from Vanuatu (South Pacific), we monitored NPP, E, LUE and WUE separately for the coconut layer (subscript "c"; LAI = 3; canopy cover around 75%) and for the under-storey composed of grasses (subscript "g"; LAI = 2.7). Light interception by the coconuts (IPARc) was estimated by optical indirect techniques (LAI-2000). Evapotranspiration of the whole stand (subscript "s"), Es, was measured directly by eddy-covariance, and the contribution of the coconuts was assessed by sapflow (Tc). Light interception and evapotranspiration from the under-storey (IPARg and Eg) was estimated from the difference. We reported elsewhere that NPPc represented 75% of NPPs (amounting to 32 tDM ha-1 year-1), Tc represented 68% of Es (amounting to 950 mm year-1) and IPARc amounted to 73% of incident PAR. This partitioning results were very close to the rule-of-thumb evaluation, based on the simple observation of the canopy closeness (0.75%). We found here that WUEs (mean annual value = 3.7 gDM kgH2O -1) was mainly driven by the coconuts (4.0), and to a lesser extent by the understorey (2.4). WUEs had high seasonal variations, b