A SNR-C model to quantitatively predict the leak gas detection capability of thermal imager was proposed. Using this model, methane gas concentration of cooled thermal imager GasFindIRTM at SNR=1 was predicted, which was consistent with the measured results. The indoor testing setup was designed and built, SNR-C curves of uncooled thermal imager Photon320 to detect ethylene were tested at different temperatures of blackbody, namely 298, 303, 308, 313 and 318 K. Through analysis, measurement and prediction concentration at SNR≈5 were relatively close in all of five blackbody temperatures, and respectively prediction concentration are 3146, 987, 570, 394 and 298 ppm. And this variation trend was consistent with the measured concentration. Therefore, the SNR-C model put forward in this article was able to predict the gas detection capability of thermal imager. While the testing setup can quantitatively measure the relationship between SNR and gas concentration, which can be applied to test the indoor performance of gas leak detection of thermal imager.