In　this　paper,　we　propose　a　key　generation　scheme　based　on　face　and　apply　it　in　online　authentication.　In　order　to　improve　security　and　the　tolerance　to　intra-class　variation,　randomization　and　statistically　optimal　algorithm　are　chosen　to　generate　the　key.　In　enrollment　stage,　a　128-dimensional　principal　component　analysis　(PCA)　feature　vector　is　firstly　extracted　from　the　face　image.　And　a　randomized　feature　process　is　utilized　to　improve　the　security　and　control　the　intraclass　variations　of　biometric　data　to　the　minimal　level.　From　the　binary　vector　of　128　bits　we　select　the　statistically　distinguishable　bits　to　form　bio-key.　Furthermore,　an　error-correct-code　(ECC)　is　generated　using　Reed-Solomon　algorithm.　In　authentication　stage,　the　same　procedure　is　implemented　to　extract　PCA　features　and　randomize　the　feature　vectors.　Then　a　bio-key　is　generated　using　the　look-up　table　and　auxiliary　code.　The　on-line　authentication　mainly　relies　on　checking　the　validity　of　the　biokey.　The　experimental　results　using　ORL　face　database　shows　that　our　algorithm　is　more　effective.