Flutter　derivatives　are　identified　via　the　wake-induced　vibration　of　bridge　deck　model.　The　error　index　is　proposed　to　evaluate　the　quality　of　the　identified　derivatives.　Output　covariance　is　utilized　to　obtain　the　Markov　parameters　of　the　system,　which　is　subjected　to　ERA　method,　subsequently,　for　the　extraction　of　flutter　derivatives.　The　error　index　is　established　within　the　framework　of　ERA.　The　Effect　of　nonlinearity　on　the　identification　results　is　discussed　through　the　comparison　of　the　flutter　derivatives　and　error　index　of　the　wake-induced　random　vibration　and　free　decay　vibration　cases.　It　is　shown,　by　the　experimental　results　of　two　models　with　different　sectional　shapes,　that　the　identification　error　is　clearly　related　to　the　nonlinearity　of　the　interactive　system.