In this paper, a novel method for predicting the visibility of packet losses in SD and HD H.264/AVC video sequences and modeling their impact on perceived quality is proposed. Based on the findings of a new subjective experiment it is initially shown that the classification of packet loss visibility in a binary fashion is not sufficient to model the perceptual degradations caused by the transmission errors. The proposed no-reference algorithm extracts a set of features from the video bit-stream to account for the spatial and temporal characteristics of the video content and the induced distortion due to the network impairments. Subsequently, the visibility of packet losses is predicted in a continuous fashion using Support Vector Regression. Finally, a no-reference bit-stream based video quality assessment model that explicitly employs the predicted packet loss visibility estimates is presented. The evaluation of the proposed model demonstrates that the use of continuous estimates for the visibility of packet losses improves the performance of the video quality assessment model.