The goal of this project is to link computations with network mechanisms in complex sensory processing. To achieve this we will integrate three approaches, currently used separately to study sensory coding in cortical circuits. First, experimental technology makes it possible to record and manipulate the activity of identified neurons within their local cortical environment. Second, functional models for the processing of complex stimuli have matured to a level that they can produce compact, predictive descriptions of the neuronal responses in terms of computations on sound features. Third, dynamical models of recurrent neural networks have reached a level in which they can incorporate many cortical details of the real cortex. While these approaches obviously complement each other - functional models uncover the computations that neurons perform while network models suggest the mechanisms underlying these computations - they are rarely combined. Here we will integrate the three approaches in order to create a spiking network model whose neurons perform computations mimicking those of real cortical neurons. Choosing primary auditory cortex (A1) as a test case, we will use the responses to pure tones to constrain a network model operating in the "balanced excitation/inhibition regime." We will then use the responses of real A1 neurons to dynamical stimuli called "tone clouds: in order to estimate functional models that extract the features to which A1 neurons respond. We will apply the same functional models to extract features to which model neurons in the spiking network responds. These comparisons will make it possible to uncover the similarity and differences between the computations carried out by the real and by the model neurons, and to create a spiking network model of auditory cortex that is functionally correct. Successful completion of the project will provide an understanding of the computational significance of anatomical and physiological components of cortical networks in auditory processing. It will also provide predictions regarding the role of inhibitory networks in the processing of complex sounds. These predictions will be tested using optogenetic manipulations.