Synthesized and directly acquired spin-echo images were compared in order to assess the validity of magnetic resonance (MR) image synthesis as a method enabling retrospective formation of images by interactive manipulation of scan parameters. Synthetic images subjectively compared favorably in both accuracy and precision with acquired images when formed for the same values of echo (TE) and repetition times (TR) and for interpolated and extrapolated values of both TE and TR. Plots of synthetic and acquired signals within the same pixel sectors quantitatively showed comparable values for several regions of interest in the brain. Percent error and noise-normalized differences between acquired and synthetic images were tested as a quantitative measure of accuracy. Percent error was consistently less than 5% for brain parenchyma, and synthetic signals were accurate to within four times the noise level at acquisition. The apparent signal-to-noise ratio of synthetic images was comparable, superior, or inferior to similar acquired images, depending on the values of TE and TR. Total acquisition time required for synthetic formation of images for arbitrary values of TE and TR was equivalent to that of a single direct acquisition with a TR of 2,500 msec.