Infrared upconversion offers advantages over direct detection of thermal imagery, but has been of limited use because of low upconversion efficiencies and because of the complexity of the upconversion systems. Photon‐limited imaging techniques can overcome the disadvantages of upconversion. The low‐power levels required by photon‐counting detectors make large upconversion efficiencies unnecessary. At the same time, photon‐limited imaging offers advantages in speed of operation and ease of implementation. We investigate the application of photon‐limited imaging to sum‐frequency upconverters, infrared quantum counters in alkali‐metal vapor, and a recently reported infrared phosphor. Upconversion efficiencies and noise effects associated with the different upconversion methods are derived. Two figures of merit are used to compare the upconversion methods. One figure of merit is the conventional noise equivalent differential temperature (NE&Dgr;T). The other is a criterion based on the statistics of photon‐limited images applied to scene matching. We find that the sum‐frequency upconverter and the phosphor require fewer detected photons than the infrared quantum counter to achieve the same values for the figures of merit. We also find that reliable scene matching can be performed with fewer detected photons than would be expected from the NE&Dgr;T figure of merit.