Seasonal re-forecast ensembles with the CCSM3.0 coupled climate model are presented, and the mid-latitude winter dynamics analyzed. The forecasts were carried out using observed ocean, atmosphere and land initial states for 1 January for 20 years. For each year, a 10-member ensemble (atmospheric resolution of T85) was integrated for one calendar year. We focus on the role of circulation regimes in the Pacific North-American region in linking intra-seasonal to seasonal mean variability and predictability. The regimes are identified on the basis of “low-frequency” (periods greater than 10 days) planetary wave variability coupled to low-frequency storm track variability. The realism of the forecasts are assessed vis-à-vis analyses, with particular focus on the significance of the regimes, their coupling to storm track variations, and their association with seasonal mean tropical Pacific SST anomalies. The design of the ensembles allows for a study of perfect model error growth from small initial atmospheric state differences, and in particular the short-time predictability of the regimes.