This paper demonstrates that improved forecasts of the annual number of hurricanes in the Atlantic tropical basin are possible by separating tropical-only hurricanes from hurricanes influenced by extratropical factors. It is revealed that variables previously shown to have a predictive relationship with the annual number of Atlantic hurricanes have a significantly stronger predictive association with the number of hurricanes formed solely from tropical mechanisms. This stronger relationship exists for extended-range (6-month lead) as well as for short-range (0-month lead) forecast models. Any future study of seasonal hurricane activity over this region should consider tropical-only hurricanes as separate from hurricanes influenced by baroclinic disturbances. The annual number of hurricanes that form or intensify as a result of interactions with baroclinic disturbances appears unrelated to significant tropical or midlatitude atmospheric anomalies and thus should be considered the random component of seasonal hurricane activity, at least until further insights are gained. Indeed, when prediction algorithms are developed to forecast the annual number of Atlantic hurricanes, best hindcast skill results from models that assume a simple average for baroclinically influenced storms. These regression-based forecast models are only marginally better than climatology.