In　order　to　detect　muscle　fatigue　effectively,　we　recorded　surface　electromyographic　(sEMG)　signals　on　the　right　upper　limbs　of　ten　young　men　while　they　were　implementing　handgrip　tasks.　Wavelet　packet　transform　and　back　propagation　neural　network　were　designed　to　extract　features　of　sEMG　and　recognize　the　muscle　states.　7-fold　cross-validation　was　used　to　test　the　results.　Our　results　showed　a　very　efficient　fatigue　recognition　using　these　methods　even　if　a　larger　scale　analysis　would　have　been　better.　The　study　indicates　that　muscle　fatigue　could　be　detected　by　analyzing　the　sEMG　signals,　which　allow　us　to　consider　a　promising　future　for　practical　applications.　©　2011　IEEE.