Emission computed tomography with a rotating camera places stringent requirements on camera uniformity and the stability of camera response. In terms of clinical tomographic imaging, the statistical accuracy required for camera flood correction, the requirements for flood accuracy, the utility and validity of flood and data image smoothing to reduce random noise effects, and the magnitude and effect of camera variations as a function of angular position, energy window and tuning were studied. Uniformity of the corrected flood response must be held to better than 1% to eliminate image artifacts that are apparent in a million-count image of a liver slice. This requires calibration with an accurate, well-mixed flood source. Both random fluctuations and variations in camera response with rotation must be kept below 1%. To meet the statistical limit, at least 30 million counts are required for the flood-correction image. Smoothing the flood image alone introduces unacceptable image artifacts. Smoothing both the flood image and data appears to be a good approach toward reducing noise effects. Careful camera tuning and magnetic shield design provide camera stability suitable for present clinical applications.