The new algorithm presented in this work allows the complete induction of neural networks encoded in linear chromosomes of fixed length (the genotype) which, nonetheless, allow the evolution of neural networks of different sizes and shapes (the phenotype). Both the chromosomal organization and the genetic operators especially developed to evolve neural networks allow an unconstrained search throughout the solution space as any modification made in the genotype always results in valid phenotypes. Furthermore, as shown for the 6-multiplexer problem presented in this work, the multigenic nature of GEP-nets can be further explored to evolve complex neural networks with multiple outputs.