The assembly of deformable components is difficult to automate due to the complex assembly movements and uncertain component geometry. Sensor-guided or deliberately compliant robots can address these issues at the cost of an enlarged parameter space of the assembly process. Machine Learning in the form of artificial neural networks (ANN) is proposed as an approach to find suitable parameter sets for such an assembly process. In this paper the automated assembly of automotive wheel arch liners by a force/torque controlled robot is considered. Two applications of ANNs are investigated experimentally: the parametrization of the assembly robot’s trajectory and the estimation of assembly offsets based on torque- and position data.