We have developed methodologies for parameter estimation and sensitivity analysis of Land Surface Models (LSMs) both in coupled and uncoupled modes. The procedures involve the use of highly efficient global multi-criteria optimization and statistical algorithms that allow for the use of the information content from different streams of observational data, e.g. fluxes -sensible, latent heat, and CO2 together with state variables -soil moisture and ground temperatures. The techniques were applied to seven different land surface models covering different level of complexity in the representation of the physical processes: BATS, BATS2, NOAH, CLM, LSM, CHASM, and Bucket. Data from more than 10 different locations covering different climates, soils and vegetation types have been used. We have found that the use of the multi-criteria procedures provide a number of advantages over the traditional single-criterion approach such as: constrain the model outputs and the parameters to physically meaningful values; obtain areal estimations of the parameters; give insight into model structure and code deficiencies; and model performance evaluation/comparison. The procedures allow for a reduction in the model overall error in the order of 20 to 50%. We have also studied the influence of the feedback effects on the parameter values using LSM coupled to the NCAR Single Column Model.