Because it is impossible to obtain real-image data that represent all of the scenarios and environmental conditions under which target recognizer models should be tested, we have developed a scene composer to aid in the analysis and evaluation of multiple-sensor electronic vision systems. The scene composer can present real-image data, computer-generated synthetic image data, and/or data composed of both real and synthetic imagery merged together so as to mimic an actual scene. For example, synthetic targets can be merged with real-image background scenes or vice versa. In addition, the composer uses a simple multiple scattering model to simulate image degradations due to fog or dust. Presently, our synthetic generator is a simple infrared scene model based on the assumption that a given absolute temperature in a scene will be detected as a graybody radiance value. The temperature model allows wavelength-dependent functions such as surface emissivity, detector responsivity, and spectral filter characteristics to be included in the integration of the Planck equation. Real-image data have been obtained from the Texas Instruments 8-bit LANTIRN Database and from other sources. The paper includes examples of real, synthetic, and merged infrared images and images degraded by simulated fog and dust.