In this study we present the ICON model with strongly improved simulation of El Niño Southern Oscillation (ENSO). Starting from ENSO performance below the average of state-of-the-art (CMIP) climate models, the model now performs at a level comparable to, or exceeding, that of the best published CMIP models. We achieved this by a series of artificial corrections. First, we introduced flux adjustments of the net heat flux and wind stress to correct biases in the tropical mean sea surface temperature (SST) and mean momentum flux. We further optimised a set of atmospheric tuning parameters to minimise a cost function based on ENSO metrics. This has led to substantial improvements in large-scale SST amplitude, pattern, seasonality, non-linear behaviour, Bjerknes and cloud feedbacks. All three elements of our approach have contributed to improved ENSO simulations, but the largest improvements come from the heat flux adjustment that reduces mean SST biases. The tropical flux adjustment also improved tropical mean precipitation patterns but had only minor improvements in the mean climate outside the tropics. The tuning of atmospheric model parameters to improve ENSO shows negative side effects on the global mean climate, which were not part of our optimisation cost function. The results suggest that flux adjustments of the tropical mean state can be beneficial for ENSO variability and should help to improve seasonal to decadal forecasting. Tuning of model parameters is more problematic, can lead to unwanted side effects and may require substantially larger computational resources than considered in this study.