This paper investigates model performance and parameter behavior for a range of land surface model (LSM) complexity across a variety of vegetated surfaces. Although LSMs are used routinely in regional and global climate (and weather) prediction, there has been limited rigorous testing of these models across a range of biomes. A systems-based approach is used to compare five commonly used LSMs (BUCKET, CHASM, BATS1e, BATS2, and Noah) across five different vegetated sites (pasture, short grass, cropland, tropical rain forest, and semiarid desert). Results indicate that there is no "perfect" model and that additional complexity (defined as additional physical representation in a model) does not necessarily equate to improved performance. In general, the medium complexity BATS1e model has the most consistent performance, with overall lower errors across the sites. Results also indicate that prescribed parameter meanings are not consistent across the various LSM formulations. A comparison of BATS1e and BATS2 parameters reveals significant differences in behavior across the study sites. These findings have key implications for general application of a single model across a range of global biomes and for model intercomparison studies where parameters are preassigned to participating models.