The investigation of imitation, which consists of observation and later reproduction of voluntary actions, promises insights into the complex processes of human actions. Although several aspects concerning the component neural processes necessary for action execution are known, our current understanding of the neural networks underlying these remains sparse. The present study applies independent component analysis (ICA) to functional magnetic resonance imaging (fMRI) data acquired during imitation of abstract gestures and object-related actions. This enables identification of neural networks underlying the production of these imitations. The explorative approach of ICA is complemented by an analysis of time courses from the maxima of each component. Four independent networks were active during delayed imitation. These can be assigned to the aspects of (1) action perception, (2) motor preparation and action execution, (3) encoding and retrieval into and from working memory, as well as (4) the dynamic integration of object affordances into the action. At least two of these networks participate in action preparation, one contains areas involved with motor working memory and one includes areas which are connected to the true action execution. The fourth network only shows activity shortly before an object-related action is imitated. This indicates a late integration of object affordances into the movement as the time course of activity in this network pertains to action rather than perception of the object. Hum Brain Mapp, 2009. © 2009 Wiley-Liss, Inc.