There is currently limited understanding of the mechanisms of decadal climate variability, and of the potential predictability of the climate system on decadal time scales. Models currently used for decadal and longer climate change projections do not start their projections from the observed state of the ocean. Therefore, a potential source of skill for decadal climate change simulations is neglected. On the decadal scale, the relative roles of forced climate change and internal natural variability may be comparable. Thus, an improved understanding of decadal variability and predictability could lead to significant improvements of decadal scale climate projections. One potentially important region is the Atlantic, where multi-decadal scale warming has apparently led to increased hurricane activity. The relative contributions of anthropogenic forcing and internal variability to that increase of hurricanes is unknown, but it is precisely this question that is crucial for future estimates of hurricane activity. We describe a systematic program of research activities whose aim is to (i) improve our understanding of the mechanisms of Atlantic decadal variability, (ii) evaluate potential predictability of the climate system, (iii) develop the necessary tools to make decadal climate predictions starting from observed ocean states, and (iv) conduct ensembles of decadal climate predictions starting from estimates of the observed state of the ocean. This research will be primarily conducted using GFDL s CM2.1 global climate model, as well as future climate models currently under development at GFDL. A crucial component of the research will be the further development and use of a novel assimilation technique recently developed at GFDL. The outcome of the research should be (i) an improved understanding of the mechanisms of Atlantic decadal variability, (ii) an evaluation of decadal scale predictability, (iii) a prototype system for making decadal climate predictions, including a newly developed assimilation system that will make state of the art estimates of the ocean from modern observational networks, and (iv) several ensembles of experimental decadal scale forecasts.