Inferring AS-level end-to-end paths can be a valuable tool for both network operators and researchers. A widely known technique for inferring end-to-end paths is to perform traceroute from sources to destinations. Unfortunately, traceroute requires the access to source machines and is resource consuming. In this paper, we propose two algorithms for AS-level end-to-end path inference. The key idea of our algorithm is to exploit the AS paths appeared in BGP routing tables and infer AS paths based on the ones. In addition, our algorithms infer AS paths on the granularity of destination prefix instead of destination AS. That is, we infer AS paths from any source AS to any destination prefix. This is essential since routing in the Internet is determined based on destination prefixes instead of destination ASs. The validation results show that our algorithm yields accuracy up to 95% for exact match and accuracy up to 97% for path length match. We further extend our algorithm to infer a set of potential AS paths between a source AS and a destination prefix. We find that on average, 86% of inferred AS path sets are accurate in the sense that one of the paths in the set matches the actual AS path. Note that our algorithms require BGP routing tables only and do not require additional data trace or access to either sources or destinations. In addition, we demonstrate that the accuracy of this BGP-based inference approach cannot go beyond 90%.