In this work we discuss the parallelization of the model selection process for Tree Echo State Networks. We consider two different “structured” parallelization strategies: one based on functional replication of the computations needed to evaluate the different steps in the model selection process, and the other one exposing and exploiting the dependency graph in the aggregate selection process computations. Both parallelizations have been implemented using FastFlow. Experimental results on state-of-the-art multicore architectures are discussed in detail that demonstrate the feasibility and the efficiency of the parallelizations.