An adaptive-parameter recursive least square(RLS) algorithm for adaptively learning connection weights and parameters of activation function was presented for multi-target rotation-invariant classification.Computer simulations reveal that,both the learning speed and the recognition rate are improved.