Neurons in the cat superior colliculus (SC) can integrate information from different sensory modalities and enhance their responses to cross-modal stimuli in spatiotemporal coincidence. These SC neurons receive unisensory inputs from many subcortical and cortical areas, but inputs from association cortex are critical. They are essential for the development of multisensory integration and for its expression during adulthood. The mechanisms underlying multisensory integration can be clarified with the use of mathematical models and computer simulations, and in recent years, we proposed a neural network model of the SC that can reproduce the different experimental results such as multisensory enhancement, and cross-modal and within-modal suppression. However, the model was unable to account for the maturation of multisensory integration or for its loss in adulthood during cortical deactivation. The objective of this work is to present an improved model which is able to explain these physiological features of multisensory integration and which incorporates recent neurological observations concerning the convergence patterns from cortical and subcortical sources and the impact of specific receptors. The model assumes the presence of competitive mechanisms between the inputs and nonlinearities in the NMDA receptor response. In a first stage of computer simulations, synaptic weights are arranged to mimic the behavior of an adult SC neuron, and the simulation produces results comparable to those from empirical physiological studies, including those with cortex deactivated and with NMDA receptor blockade. In a subsequent stage, development is simulated by assuming that association cortico-collicular synapses are present but not active, so that SC activity is dependent completely on other sensory inputs. Sensory experience is modelled by a “training phase� in which the network is repeatedly exposed to modality-specific and cross-modal stimuli at different locations in space. The synaptic weights of association cortico-collicular synapses are modified based on Hebbian rules. After the training period, SC neurons respond in an adult-like fashion to modality-specific and cross-modal stimuli. The model summarizes much of the present knowledge regarding cat SC and, using realistic synaptic learning rules, provides a theoretical framework for understanding how the development of multisensory integration is guided by early sensory experience.