We assess the seasonal forecast skill of pan-Arctic sea ice area in a dynamical forecast system that includes interactive atmosphere, ocean, and sea ice components. Forecast skill is quantified by the correlation skill score computed from 12 month ensemble forecasts initialized in each month between January 1979 to December 2009. We find that forecast skill is substantial for all lead times and predicted seasons except spring but is mainly due to the strong downward trend in observations for lead times of about 4 months and longer. Skill is higher when evaluated against an observation-based dataset with larger trends. The forecast skill when linear trends are removed from the forecasts and verifying observations is small and generally not statistically significant at lead times greater than 2 to 3 months, except for January/February when forecast skill is moderately high up to an 11 month lead time. For short lead times, high trend-independent forecast skill is found for October, while low skill is found for November/December. This is consistent with the seasonal variation of observed lag correlations. For most predicted months and lead times, trend-independent forecast skill exceeds that of an anomaly persistence forecast, highlighting the potential for dynamical forecast systems to provide valuable seasonal predictions of Arctic sea ice.