We introduce an illumination-invariant target color detection technique for EMCCD camera. The EMCCD camera has an advantage to acquire a color image even though the surroundings are very low-light-levels. Since the EMCCD amplifies the input signal, an EMCCD noise which is caused by the gain process in low-light imaging systems occurs. In order to use the EMCCD image, we firstly reduce the noise by using selective Gaussian blur filter which maintains the image boundary. The threshold for the Gaussian blur filter is decided after the noise analysis by using sample patch modeling. We use illumination-invariant color space based on the image acquisition model for the target color detection. Experimental results show a good possibility to apply to intelligent surveillance system and night vision system.