We developed a novel deconvolution software (NeuroDeblur), which significantly enhances the visibility of minute details, such as neurons or nerve fibres, in large image stacks captured by light-sheet, confocal, or bright-field microscopy. NeuroDeblur employs synthetic point spread functions tailored to each of these imaging modalities by utilizing a sophisticated optical model of the imaging procesess. Experimental point spread functions derived from recordings of fluorescent beads can be used additionally. The software features a sophisticated adaptive background removal algorithm and automatic block-wise processing of large data sets, enabling the handling of very large image stacks (>100 GB) even on modest computers equipped with 8 or 16 GB RAM. Through extensive parallelization and optional GPU acceleration, NeuroDeblur achieves high-speed performance. For instance, on a PC equipped with a state-of-the-art NVidia graphics board, a 3D-stack comprising approximately 1 billion voxels can be deconvolved within 5-10 minutes.