Using a factorial experiment framework, a systematic relationship is developed between uncoupled and coupled land surface model sensitivities. It is shown that the relative sensitivity of a coupled land-atmosphere model can be partitioned into the relative sensitivity of the corresponding uncoupled model, a term related to interactive effects, and a product term involving main effects of land and atmospheric processes. A widely used land surface model is used in its uncoupled and coupled modes to illustrate the methodology. The proposed methodology demonstrates the need and utility to combine the traditional one-factor-at-a-time method (e.g., relative sensitivity analysis) with a factorial experiment framework. The traditional sensitivity analysis cannot evaluate the multifactor effects. The factorial design experiments, on the other hand, can estimate multifactor effects but are not dimensionless and hence depend on the chosen range of parameters. The proposed method offers the best of both approaches by estimating multifactor effects with dimensionless quantities. The proposed framework is used to explain why sensitivities to certain land surface parameters are enhanced or damped when the land surface module is coupled with an atmospheric model. For the tested cases there does not appear to be any systematic trend in sensitivities of the coupled and uncoupled models. It is argued and shown that similarity of relative sensitivities between an uncoupled and coupled model does not necessarily imply that the influence of land-atmosphere feedback is negligible. By decomposing the relative sensitivity of a coupled model, we have shown that interactions between land and atmosphere cannot be neglected while testing land surface representations. ¿ American Geophysical Union 1996