A number of computer programs exist for the determination of ejection fraction from gated blood pool images. Three commercially available algorithms were compared on the same radionuclide data from 50 catheterized patients. The methods used were: (1) simple thresholding, (2) second derivative edge determination, and (3) a combination of these two. There was good correlation with catheterization ejection fraction for all three methods using smoothed data [r = 0.781, 0.77, 0.84 respectively (P less than 0.01)]. However, interobserver variance was significantly decreased (P less than 0.01) by use of the second derivative method when compared with the other two methods and catheterization. Thus, if accuracy and low interobserver variance are required, then a combination of a second derivative and thresholding method is preferable.