From natural guide star adaptive optics data, we study the properties of the on-axis point spread function and discuss the performance of adaptive optics images. A close relation between the full width half maximum and the Strehl ratio is shown. The ultimate exploitation of adaptive optics images requires the deconvolution and therefore the calibration of the point spread function. This calibration is commonly made by sequentially observing a point source close to the astronomical target. In the partial correction regime, the calibration mismatch is the main source of noise or bias in the deconvolution process and it is mainly induced by the varying seeing conditions. Complementary techniques like near-sighted deconvolution, speckle and rebinning are briefly reviewed. Based on observations made at ESO and SOR.