Noise estimation plays an important role in the evaluation of video quality. In video surveillance systems, noise is mainly introduced by the camera and the quantization process. This paper proposes a method to estimate the noise in video communication systems. Firstly, the variance of noise introduced by the camera can be estimated by using inter-frame and intra-frame differential operation. Then we construct a segmented function according to the estimated variance of camera noise, quantization step, and the DCT coefficients of each micro-block decoded from the video stream. Finally, the variance of overall noise in the decoder side will be evaluated. The simulation shows that the estimation is fairly accurate, so that this method is suitable for real noise evaluation in video surveillance systems.