Prevent undersampling (voxelsize too large, especially with objectives of high numerical aperture: results in missing signals between the physical (illuminated and recorded) sampling point (problem of aliasing)
Prevent oversampling (too short distances of scanning points do not increase resolution, but result in too large data sets and waste of time and may enhance bleaching)
Fast checks and recordings (overviews etc.)
for 63x/1.3 und 100x/1.3 objectives: choose a voxelsize X/Y: about 150-200 nm, voxelsize Z: about 200-500 nm.
Large voxel sizes may be necessary in cases where fast scanning is necessary or bleaching is a severe problem. This may be achieved (without missing structures) by lowering the numerical aperture of the objective and / or by opening the pinhole.
High resolution scans for deconvolution
Choose appropriate voxel sizes (sampling distances) in X/Y and Z direction (scan more slices and less accumulation per slice, if possible (in most cases, no accumulation is ok). Adhere to the Nyquist criterion. For calculation, see:
Set accurate offset (low background, but no zero value pixels) and PMT setting (almost no overflow channel, i.e. highest intensity pixels). Otherwise the correction by deconvolution may not be correct. In case of an offset too high (too much zero value pixels), negative noise may be filtered out and calculation may be wrong caused by incorrect distribution of gray values.
Include all visible structures plus some space beyond visible structures, because they may contain low signals needed for correct deconvolution (e.g. 3 micrometers in each direction beyond those signals which are obviously generated by the structures of interest).
Use 12 bit mode or higher (better definition of weak signals / structures in high dynamic range scans)