Leaf Area Index (LAI), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) and the fraction of vegetation cover (FCOVER) are key variables to describe and monitor the status and phenology of crops, and are related to a variety of surface processes such as evapotranspiration, photosynthetic capacity and productivity. Several studies demonstrated the capacity to derive crop biophysical variables from decametric surface reflectance data either based on physically-based methods (e.g., by inversion of PROSPECT-SAIL model and stochastic radiative transfer (RT)) or based on empirical approaches relating spectral information or 5th INTERNATIONAL SYMPOSIUM: RECENT ADVANCES IN QUANTITATIVE REMOTE SENSING 109 vegetation indices with ground observations. Previous studies are mainly based over one site or over particular crop types. However, there are still few evaluative studies that compare several retrieval methods over both generic crops or by specific crop types. The objective of this study is to assess the performance of physically-based and empirically-based methods to retrieve biophysical variables over either generic crops or specific crop types using Landsat-8 imagery. The following methods are considered: inversion of PROSPECT+SAIL using neural networks, stochastic RT and several empirical approaches based on machine learning algorithms (i.e., ordinary least squares, neural networks, gaussian processes, support vector machines) which were trained either using the several Landsat-8/OLI spectral bands or using selected vegetation indices. We used the FP7 ImagineS ground database (LAI, FAPAR and FCOVER) with more than 2000 ESUs (Elementary Sampling Units) corresponding to various crops types and conditions over 20 different sites from 2013 to 2016. Finally, 900 ESUs were collocated with clear-sky Landsat-8/OLI top of canopy multispectral reflectance values and corresponding sun zenith angle. Five main crop classes were identified: Rice, Wheat & Barley, Corn, Sunflower, Soybean and mixed vegetables (including other minor types like Potato, Alfalfa, Pepper, Tobacco, Sugarbeet and Onion). Three subsets have been created, two for training of algorithms and one for the validation. The three possible combinations of training and validation were executed in the exercise. The methods under study were evaluated on each crop type and for all together. As a result, scatter-plots and a summary matrix of the validation statistics for each retrieval technique and each crop type are presented and discussed.