Ocean surface wave height in the Atlantic Ocean is strongly influenced by the North Atlantic Oscillation (NAO). Here we demonstrate for the first time a skilful seasonal forecast for wave height in the Atlantic Ocean, produced by a seasonal prediction system with an enhanced prediction skill of winter NAO. The improved seasonal prediction skill of the wave height reaches 0.8 in major parts of the North Atlantic. Prediction skill in the Central and South Atlantic is significantly improved due to swell propagation from better represented active wave generation regions in the North Atlantic. By subsampling, the modeling of climatological anomalies of seasonal wave height for strongly positive and negative NAO phases is considerably improved. We demonstrate the potential of an improved, subsampling‐based approach for the dynamical seasonal prediction of waves, specifically for extreme seasons during strong NAO phases, which can be implemented for operational purposes. Plain Language Summary Ocean surface wave height in the Atlantic Ocean depends mainly on low‐frequency atmospheric variability such as the North Atlantic Oscillation (NAO). Depending on the NAO phase, different weather regimes, mean, and extreme wind and wave conditions develop over the North Atlantic. The NAO affects the location and orientation of cyclone tracks and is therefore responsible for more frequent extreme storms during a strongly positive NAO phase. Here for the first time, we show that a state‐of‐the‐art seasonal prediction system with an enhanced prediction of winter NAO leads to better forecasting of ocean waves in the Atlantic Ocean. In major parts of the North Atlantic, the classical ensemble mean approach demonstrated a prediction skill for the seasonal mean wave height of less than 0.5 for the hindcast period from 1982 to 2017. In contrast, the ensemble subsampling approach increased the skill to up to 0.8. Modeling of the seasonal mean wave height for strongly positive and negative NAO phases is considerably improved after subsampling. Thereby, we demonstrate the potential of a subsampling approach for the prediction of wave conditions during strong NAO phases.