The dendrites of cortical pyramidal neurons receive synaptic inputs from different pathways that are organised according to their laminar target. This architectural scheme provides cortical neurons with a spatial mechanism to separate information, which may support neural flexibility required during learning. Here, we investigated layer-specific plasticity of sensory encoding following learning by recording from two different dendritic compartments, tuft and basal dendrites, of layer 2/3 (L2/3) pyramidal neurons in the auditory cortex of mice. Following auditory fear conditioning, auditory-evoked Ca2+ responses were enhanced in tuft, but not basal, dendrites leading to increased somatic action potential output. This is in direct contrast to the long held (and debated) hypothesis that, despite extensive dendritic arbours, neurons function as a simple one-compartment model. Two computational models of varying complexity based on the experimental data illustrated that this learning-related increase of auditory responses in tuft dendrites can account for the changes in somatic output. Taken together, we illustrate that neurons do not function as a single compartment, and dendritic compartmentalisation of learning-related plasticity may act to increase the computational power of pyramidal neurons.Significance StatementThis study directly investigates whether information processing in neurons is compartmentalized within different dendritic regions. Our findings shed light on the learning-related changes that occur in dendritic compartments residing in different layers of the cortex (i.e. tuft and basal dendrites), illustrating that L2/3 pyramidal neurons are able to compartmentalize learning signals which leads to enhanced somatic output. The compartmentalization of experience-dependent plasticity supports flexible sensory processing and may increase the computational power of single pyramidal neurons. In addition, it highlights the special role of tuft dendrites, the target of top-down inputs, in modulating sensory encoding following fear learning.