A source coding distortion estimation scheme using one-pass encoding is proposed to predict the distortion in the next frame based on the distortion of the last coded frames. Motion-compensated prediction is analyzed first, in the case that quantization step sizes used for different video frames may be different due to bit-rate constraint in constant bit-rate real-time applications. Second, the proposed scheme uses a linear prediction model to compute the variances of the discrete cosine transform (DCT) coefficients. Third, the DCT coefficients are assumed to follow a Laplacian distribution and the distortion model of the Laplacian source is used to compute the coding distortions of the DCT coefficients. Fourth, the proposed scheme calculates a distortion error which appears because of assuming the direct current coefficient to be Laplacian distributed instead of Gaussian distributed. Finally, the source coding distortion in the frame is predicted using the coding distortions in the DCT coefficients and the calculated distortion error. The proposed scheme does not need trial encoding and can be applied to real-time video coding. When compared to the other five existing schemes, experimental results show that the proposed scheme can estimate source coding distortion more accurately than the five schemes.