RNA(ribonucleic acid) sequence feature extraction is widely used in various research fields related to RNA analysis, such as gene expression analysis, RNA structure prediction or interaction study. Considering that an RNA sequence is composed of four nucleotides, quaternions were introduced into the feature extraction of RNA sequences for the first time. Four new feature extraction methods were proposed, which were the pure quaternion coding method, the unidirectional superposition method, the center bidirectional superposition method, and the Fourier transform method. Two of these methods combined both the whole and local frequency and position information of the sequence. Four new methods were proposed to convert RNA sequences into quaternion vectors of the same dimension for feature extraction. KNN (K-nearest neighbor) classifier and the SVM (support vector machine) classifier were compared with common feature extraction methods such as NAC (nucleic acid composition), ANF (accumulated nucleotide frequency) and TNC (tri-nucleotide composition). The results show that the proposed methods have good performance.