In this paper, we propose a novel online ϵ-approximation algorithm, called LC-CloStream, for mining closed frequent itemsets embedded in a transactional stream. LC-CloStream is based on an incremental/cumulative intersection method and ϵ-elimination proposed by Lossy Counting algorithm. We show, LC-CloStream is essentially incomplete, but is still semi-complete for mining frequent closed itemsets in a stream. Moreover, we prove the completeness of extracting frequent itemsets and the ϵ-approximation for estimating the frequency. We also show several good performances of the experimental evaluation for LC-CloStream.