Even under spontaneous conditions and in the absence of changing environmental demands, awake animals alternate between moments of increased alertness and moments of torpor or disengagement. These changes in brain state can occur rapidly, on a timescale of seconds, and may be correlated with overt changes in exploratory behaviors (walking and whisking) or they may be more covert, with no external correlates except for changes in pupil size. Neuromodulators such as acetylcholine (ACh) are thought to play an important role in driving these spontaneous state transitions, and cholinergic activity in cortex has been monitored via calcium imaging of cholinergic axons and with new genetically-encoded fluorescent neuromodulator sensors. Here, we perform the first simultaneous imaging of sensors and axons in vivo, to examine the spatiotemporal properties of cortical acetylcholine (ACh) release during spontaneous changes in behavioral state. As has been previously reported, periods of locomotion were accompanied by large increases in ACh levels across the dorsal cortex, and pupil size tracked smaller, more rapid changes in ACh during periods of quiescence. We observed a high correlation between simultaneously-recorded basal forebrain axon activity and neuromodulator sensor fluorescence. Consistent with volume transmission of ACh, increases in axon activity resulted in increases in local ACh levels that fell off with the distance from the nearest axon. GRAB-ACh fluorescence could be accurately predicted from axonal activity alone, providing the first validation that neuromodulator axon activity is a reliable proxy for nearby neuromodulator levels. To more precisely understand the temporal kinetics of ACh, we applied a deconvolution approach to account for the kinetics of the ACh sensor. Deconvolution of fluorescence traces emphasized the rapid clearance of ACh, especially for smaller transients outside of running periods. Finally, we trained a predictive model of ACh fluctuations from the combination of pupil size and running speed; this model performed better than using either variable alone, and generalized well to unseen data. Overall, these results contribute to a growing understanding of the precise timing and spatial characteristics of cortical ACh during fast brain state transitions under spontaneous conditions.