Reproducibility or precision was measured for a current positron emission tomographic (PET) system and was simulated by both computer and experimental methods for potential future systems. The present and potential future precision of PET systems is on the order of 1 to 2% for equilibrium studies (such as fluorodeoxyglucose metabolism measurements) with a few million counts per image, and better than 10% for dynamic studies with a few hundred thousand counts per image. To take advantage of this high precision, the instrumental accuracy of PET systems should match or better the precision of the measurements. The major sources of inaccuracy in PET, outside of deficiencies in electronic or computer hardware and software, are poor resolution, nonuniform resolution, and scatter. The use of narrow bismuth germanate detectors, adequately large detector ring diameters, and properly designed interplane septa will provide substantial improvements in all of these deficiencies.