Except for blind methods, deconvolution of 3-D data sets acquired from a ¯uorescence microscope requires the knowledge of the point spread function (PSF) of the instrument. Using the XCOSM XCOSM package, we show ®rst with simulations and then with recorded data that it is possible to recover from an experimental PSF some parameters, which are very dicult or impossible to measure during the acquisition, like the specimen depth or the immersion medium re-fractive index. Doing so, we can precise the acquisition protocol, which helps to use the instrument under optimal conditions. Furthermore, the knowledge of the actual acquisition conditions permits to use for the deconvolution process a computed PSF, which is noiseless and as close as possible to the actual PSF. This helps to reduce errors in quantitative measurements after deconvolution, as shown with computations.