In this work, a novel approach utilizing feature covariance matrices is proposed for time series classification. In order to adapt the feature covariance matrices for time series classification, a feature vector is defined for each point in a time series. The feature vector comprises local and global information such as value, derivative, rank, deviation from the mean, time index of the point and cumulative sum up to the point. Instead of representing the whole time series with a single covariance matrix, time series is divided into overlapping subsequences. Extracted feature vectors for the time instances are concatenated to construct feature matrices for the overlapping subsequences. Covariance of the feature matrices are used to describe the subsequences. After the determination of feature covariance matrices for both training and test samples, SVM classifier is utilized to decide the class of the test samples. Conducted experiments on UCR time series dataset show that the proposed method yields results which mostly outperform well-known methods such as DTW, shapelets and other state-of-the-art techniques.