Feature coding plays an important role in image classification, and the soft-assignment coding (SAC) method is popular in many practical applications due to its conceptual simplicity and computational efficiency. The SAC method, however, fails to achieve the optimal image classification performance when compared with the recently developed coding methods, such as the sparse coding and the localityconstrained linear coding methods. This chapter first analyzes the SAC method from the perspective of kernel density estimation, and then presents an improved softassignment coding (ISAC) method that enhances the image classification performance of the SAC method and keeps its simplicity and efficiency. Specifically, the ISAC method introduces two enhancements, namely, the thresholding normalized visual word plausibility (TNVWP) and the power transformation method. These improvements are further shown to establish the connection between the proposed ISAC method and the Vector of Locally Aggregated Descriptors (VLAD) coding method. Experiments on four representative datasets (the UIUC sports event dataset, the scene 15 dataset, the Caltech 101 dataset, and the Caltech 256 dataset) show that the proposed ISACmethod achieves competitive results to and even better results than some popular image classification methods without sacrificing much computational efficiency.