A general purpose implementation of the Tabu Search metaheuristic, called U
niversal Tabu Search, is used to optimally design a Locally Recurrent Neura
l Network architecture. Indeed, the design of a neural network is a tedious
and time consuming trial and error operation that leads to structures whos
e optimality is not guaranteed. In this paper, the problem of choosing the
number of hidden neurons and the number of taps and delays in the FIR and I
IR network synapses is formalised as an optimisation problem whose cost fun
ction to be minimised is the network error calculated on a validation data
set. The performances of the proposed approach have been tested on the desi
gn problem of a Neural Network controller of a Custom Power protection devi
ce.