The performance of a combinatorial simultaneous optimisation algorithm (SO) is tested using experimental LEED I(E) data from Cu(100) and Fe$_{0.57}$Al$_{0.47}$(100) surfaces. SO optimises structures taking advantage of the experimental database at two levels: (i) conmensurate subsets of the database with the number of unknown parameters are chosen to find local solutions using Broyden's method and, (ii) these partial structural solutions are used to build a Markov chain over the whole database. This procedure is of global character, the same as simulated annealing or genetic algorithms methods, but displays a very competitive scaling law because after the first iteration candidates are not chosen by a blind/random pick; they are already solutions to the problem with a restricted experimental database.Financed by CYCIT (MAT-2005-3866) and MEC (CONSOLIDER NANOSELEC-26400)