This paper examines the self-similarity of World Wide Web (WWW) traffic, a phenomenon where traffic patterns exhibit similar characteristics at different time scales. The authors use extensive traces of NCSA Mosaic user executions to analyze the dependence structure of WWW traffic, finding evidence that it exhibits behavior consistent with self-similar traffic models. They attribute this self-similarity to the underlying distributions of WWW document sizes, the effects of caching, user preferences in file transfer, and the superimposition of many such transfers in a local area network. The study uses empirical distributions from their traces and data from over thirty WWW sites to support these findings. The paper also discusses the statistical methods used to test for self-similarity and the implications of these findings for network design and performance optimization.This paper examines the self-similarity of World Wide Web (WWW) traffic, a phenomenon where traffic patterns exhibit similar characteristics at different time scales. The authors use extensive traces of NCSA Mosaic user executions to analyze the dependence structure of WWW traffic, finding evidence that it exhibits behavior consistent with self-similar traffic models. They attribute this self-similarity to the underlying distributions of WWW document sizes, the effects of caching, user preferences in file transfer, and the superimposition of many such transfers in a local area network. The study uses empirical distributions from their traces and data from over thirty WWW sites to support these findings. The paper also discusses the statistical methods used to test for self-similarity and the implications of these findings for network design and performance optimization.