There are several heuristic optimisation techniques used for numeric optimisation problems such as genetic algorithms, neural networks, simulated annealing, ant colony and tabu search algorithms. Tabu search is a quite promising search technique for non-linear numeric problems, especially for the problems where an optimal solution must be determined on-line. However, the converging speed of the basic tabu search to the global optimum is the initial solution dependent since it is a form of iterative search. In order to overcome this drawback of basic tabu search, this paper proposes a new parallel model for the tabu search based on the crossover operator of genetic algorithms. After the performance of the proposed model was evaluated for the well-known numeric test problems, it is applied to training recurrent neural networks to identify linear and non-linear dynamic plants and the results are discussed. (C) 2004 Elsevier Ltd. All rights reserved.