In this paper, we assess the short-term forecast error of a mesoscale primitive-equation open-ocean model, induced by uncertainties in wind forcing. Statistics calculated from an ensemble of ocean states show that temperature forecast error is strongest at the top of the ensemble-mean thermocline, as a consequence of vertical displacement of the mixed-layer base around its ensemble mean. Horizontal pattern of the temperature error in the mixed-layer is mainly explained by horizontal advection and surface heat flux fluctuations. These two mechanisms and entrainment through the mixed-layer bottom are presented as the three processes responsible for thermal forecast error growth in the modeled upper ocean.