The identification of linear B-cell epitopes is important for developing epitope-based vaccines. Recently, machine learning techniques have been used in the epitope prediction, but the existing encoding schemes usually neglected valuable discriminative information. In this paper, we proposed a novel encoding scheme which combines several groups of sequence-derived structural and physicochemical features, and support vector machine was used to construct the prediction models. When applied to the benchmark dataset, our proposed method demonstrated better results than benchmark methods. Moreover, the study indicated incorporating more discriminative features may contribute to the higher prediction performance.