Cognitive processes underlying behavior such as navigation, attention, and perception are linked to specific spatiotemporal patterns of neural activity in the neocortex. These patterns arise from synchronous synaptic activity and are often detected in particular frequency bands in the cortical field potential. However, cortical activity is highly variable on multiple timescales (milliseconds to hundreds of seconds). Identifying discrete neural events underlying patterned activity within highly dynamic cortical network fluctuations thus remains a critical challenge. Here, we develop a novel analytical method to track individual network events underlying state-dependent activity with single-cycle precision. In mouse primary visual cortex (V1), we find that events underlying activity in the γ-range (30-80Hz) are linked to strongly enhanced visual encoding by V1 neurons. The spectral and laminar profile of γ activity can be recreated by patterned optogenetic stimulation of thalamocortical terminals coming from the dorsal part of the lateral geniculate nucleus (dLGN). Conversely the inhibition of dLGN strongly suppresses γ events incidence, indicating that they are linked to thalamocortical integration of visual information coming from the retina. In behaving mice, suppressing γ events strongly impairs visual detection performance and γ events incidence increases steadily prior to visually cued behavioral responses. This relationship between γ events and response is sensory modality-specific and rapidly modulated by changes in task objectives. Thus, γ events support a selective and flexible encoding of visual information according to behavioral context, suggesting a major role for these transient patterns of cortical activity.