Understanding the interaction of the atmosphere with underlying water surfaces is crucial for many scientific endeavors such as: improving evaporation models, developing surface parameterizations for atmospheric simulations, studying the ecological implications of climate change on lakes, and understanding the local-scale atmospheric dynamics in coastal areas. However, water-atmosphere interaction has in the past received less attention than land-atmosphere interaction. The Lake-Atmosphere Turbulent EXchanges (LATEX) field measurement campaign was designed to help bridge this gap and address some of the issues listed above. The experiment took place on a platform in Lake Geneva in Switzerland over the period extending from August through October of 2006. The primary instrumentation consisted of: 1) a vertical array of four sonic anemometers and four open-path H2O/CO2 analyzers, 2) a Raman scattering fiber-optic temperature profiler having a resolution of 4-mm vertically and 0.01 deg C in temperature (3 meter range: 1 meter above the water surface and 2 meters below), and 3) a lake current profiler. Additional supporting measurements included net radiation, surface temperature, relative humidity, and wave height and speed. We present the diurnal trends of momentum, heat, and water vapor fluxes as well as surface energy budgets for the whole experimental period. We also test several evaporation models of varying complexity. We then focus on small scale turbulence (the so-called subgrid scales, SGS) over the lake which cannot be captured in numerical (large-eddy) simulations. The exact fluxes and dissipations produced by the subgrid scales are computed and then compared to modeled fluxes and dissipations based on the Smagorinsky and the non-linear SGS models. An interesting result is the very good correlation observed between SGS fluxes, dissipations and model coefficients for heat and water vapor suggesting that combining the SGS modeling for the two scalars is feasible and can reduce the computational cost of large-eddy simulations. On the other hand, results suggest that the Reynolds analogy, which assumes that momentum and scalars are transported by turbulent eddies with the same efficiency, does not hold at any scale.