Four experiments were carried out using the action span paradigm. In experiment 1 we found that well-learnt, meaningful (MF) actions were imitated better than novel, meaningless (ML) actions. In experiments 2 and 3, during the encoding of MF and ML actions, participants were required to carry out different suppression tasks. In experiment 2 we replicated the advantage of MF actions over ML actions and also found that the motor suppression shortened the action span more than the other forms of suppressions (spatial and articulatory). Action encoding and motor suppression tapping the same subsystem, temporarily holding the motor information, could explain the reduced motor span obtained in experiment 2. Two alternative explanations that could have accounted for this effect were ruled out in experiments 3 and 4. In experiment 3 we verified whether the reduction of the action span was produced by the different combination of the articulatory suppression with motor suppression or with the spatial suppression. In experiment 4, we demonstrated that the reduction was not due to the motor suppression being more difficult than the other types of suppression. The critical finding that the spans of well-learnt, MF actions are longer than those of novel, ML actions observed in experiments 1 and 2 was interpreted in terms of different processing routes engaged in the imitation of these two types of actions. MF actions can be imitated along both a semantic, indirect route and a direct route leading from the visual analysis of the action to the motor system. In contrast, the imitation of ML actions is accomplished along the direct route only.