Packet classification is a fundamental task for network devices such as edge routers, firewalls, and intrusion detection systems. Currently, most vendors use Ternary Content Addressable Memories (TCAMs) to achieve high-performance packet classification. TCAMs use parallel hardware to check all rules simultaneously. Despite their high speed, TCAMs have a problem in dealing with ranges efficiently. Many packet classification rules contain range specifications, each of which needs to be translated into multiple prefixes to store in a TCAM. Such translation may result in an exponential increase in the number of required TCAM entries. In this paper, we propose a bidirectional range extension algorithm to solve this problem. The proposed algorithm uses at most two TCAM entries to represent a range, and can be pipelined to deal with multiple range fields in a packet header. Since this algorithm assumes a non-redundant rule set, i.e., no range overlap between different rules, which can be obtained by applying our previous work on redundancy removal in TCAM using a tree representation of rules. Our experiments show a more than 75% reduction in the number of TCAM entries by applying the bidirectional range extension algorithm to realworld rule sets.