Spatial competition indices for Scots pine and birch growing in heterogeneous Caledonian forest in Glen Affric were developed. Competition indices (CIs) have frequently been used in empirical models of individual tree growth. However, the contribution of CIs to the predictive ability of growth models is often very small. The poor performance of CIs may relate to the choice of spatial descriptors and functional form of the index. To improve CI performance a novel approach utilising neural networks and simulated annealing for the optimisation of CIs was developed.