This is a proposal focusing on exploring climate predictability on decadal or longer timescales. The proposed research builds upon our currently NOAA-sponsored projects in the tropical Atlantic and Pacific Oceans. These research projects have produced a set of coupled ocean-atmosphere models with novel features that we believe are well suited for understanding predictable dynamics at decadal or longer time scales. We propose to use this hierarchy of the coupled climate models to shed light on the intricate interplay among natural modes of climate variability, anthropogenic forcing and weather noise in decadal climate predictability. Our model hierarchy includes 1) an atmospheric general circulation model (CAM3) coupled to a slab ocean (CAM3-ML), 2) an atmospheric general circulation model coupled to a reduced gravity ocean (CAM3-RGO), and 3) an atmospheric general circulation model coupled to a general circulation ocean (CAM3-MOM3). These are equipped with noise filtering algorithms capable of suppressing weather noise, including the signal-noise optimization noise filter developed by Chang et al. (2007) and the interactive coupled ensemble developed by Kirtman and Shukla (2002). These noisefiltering methodologies have proven to be extremely valuable in the understanding of the role of weather noise in ENSO dynamics and its predictability. We anticipate that the same will be true in the understanding of decadal climate predictability. Large ensembles of prediction experiments will be conducted using each of these models. The experiments will be analyzed systematically to test a set of scientific hypotheses aiming at providing insight into physical mechanisms that give rise to any decadal scale predictability – one of the major objectives of this year’s NOAA CVP program. An important concern in the design of decadal prediction experiments is the prohibitive computational cost of ensemble forecasts using a high-resolution, global coupled climate model for lead times of a decade or longer. Understanding the role of weather noise, and developing techniques to mitigate its effects, can help minimize the size of the ensembles needed for operational decadal prediction and substantially reduce the computational costs.