In this paper we study the perceptual impact of data loss on MPEG-2 video coded streams transmitted over an ATM network. This impact is measured using a perceptual quality metric based on a spatio-temporal model of the human visual system. Video streams have been transmitted on top of both new network and ATM adaptation layers which provide a robust transmission by applying per-cell sequence numbering combined with a selective Forward Error Correction (FEC) mechanism. We compare their performance against a transmission over AAL5. Results show that the proposed AAL behaves better in terms of both network performance and perceived quality of the MPEG-2 decoded sequence