Unusually high western Pacific oceanic heat content often leads to El Niño about 1 year later, while unusually low heat content leads to La Niña. Here, we investigate if El Niño Southern Oscillation (ENSO) predictability also depends on the initial state recharge, and discuss the underlying mechanisms. To that end, we use the CNRM-CM5 model, which has a reasonable representation of the main observed ENSO characteristics, asymmetries and feedbacks. Observations and a 1007-years long CNRM-CM5 simulation indicate that discharged states evolve more systematically into La Niña events than recharged states into neutral states or El Niño events. We ran 70-members ensemble experiments in a perfect-model setting, initialized in boreal fall from either recharged or discharged western Pacific heat content, sampling the full range of corresponding ENSO phases. Predictability measures based both on spread and signal-to-noise ratio confirm that discharged states yield a more predictable ENSO outcome one year later than recharged states. As expected from recharge oscillator theory, recharged states evolve into positive central Pacific sea surface temperature anomalies in boreal spring, inducing stronger and more variable Westerly Wind Event activity and a fast growth of the ensemble spread during summer and fall. This also enhances the positive wind stress feedback in fall, but the effect is offset by changes in thermocline and heat flux feedbacks. The state-dependent component of westerly wind events is thus the most likely cause for the predictability asymmetry in CNRM-CM5, although changes in the low-frequency wind stress feedback may also contribute.