Wide-field calcium imaging is often used to measure brain dynamics in behaving mice. With a large field of view and a high sampling rate, wide-field imaging can monitor activity from several distant cortical areas simultaneously, revealing cortical interactions. Interpretation of wide-field images is complicated, however, by the absorption of light by hemoglobin, which can substantially affect the measured fluorescence. One approach to separating hemodynamics and calcium signals is to use multiwavelength backscatter recordings to measure light absorption by hemoglobin. Following this approach, we develop a spatially detailed regression-based method to estimate hemodynamics. This Spatial Model is based on a linear form of the Beer–Lambert relationship but is fit at every pixel in the image and does not rely on the estimation of physical parameters. In awake mice of three transgenic lines, the Spatial Model offers improved separation of hemodynamics and changes in GCaMP fluorescence. The improvement is pronounced near blood vessels and, in contrast with the Beer–Lambert equations, can remove vascular artifacts along the sagittal midline and in general permits more accurate fluorescence-based determination of neuronal activity across the cortex. NEW & NOTEWORTHY This paper addresses a well-known and strong source of contamination in wide-field calcium-imaging data: hemodynamics. To guide researchers toward the best method to separate calcium signals from hemodynamics, we compare the performance of several methods in three commonly used mouse lines and present a novel regression model that outperforms the other techniques we consider.