Fatigue represents a critical issue in many structural applications, and wind turbines are not an exception. Their dynamic response over the years determines the turbine's lifespan, meaning that fatigue loads have a clear impact on the Cost of Energy. Since the direct experimental determination of the loading state is complex or expensive, estimations arising from general operational signals can be explored as an indirect way to acquire knowledge of fatigue loading levels.A case study based on 10-minute aeroelastic simulations of a wind turbine dynamics is used to develop a Damage Equivalent Load estimation model using operational signals (typically recorded by SCADA systems) as inputs. The focus is on both the input selection and the model configuration, seeking the combination which reaches the lowest error. Three filters and two innovative wrappers (exploration and optimization) were considered within the selection. Linear and Artificial Neural Network models were implemented and compared. Results showed performances in Damage Equivalent Load estimation below 4% in terms of Normalized Root Mean Squared Error, which is promising as compared with related work. Additional conclusions were obtained concerning appropriate Artificial Neural Network configurations (net type, architecture and training algorithm), likewise the potential contribution of a proposed genetic algorithm.