This paper provides an overview of the basic concepts, current state and future directions of uniform and nonuniform multirate filter banks and their applications. Research in the areas of multidimensional filter banks, and wavelet transforms, as a direct extension to the topic is still proceeding at a rapid rate. Different design techniques and algorithms that are currently of interest are investigated from hardware complexity and reconstructed signals' quality points of view. Important similarities and differences between different filters categories are noted, which lead to a discussion of important issues that need to be resolved. Distortions and quantization effects present in the filter banks are reviewed and some solutions for controlling different types of distortions are introduced. The problem of designing QMF banks that can provide good frequency resolution while allowing for exact or near perfect reconstruction of the signal is quite challenging, because so many dissimilar types of distortion must be minimized and/or eliminated in the same design context. Recently, the perfect reconstruction cosine-modulated filter bank and filter banks with block sampling have emerged as an attractive choice of filter bank with respect to implementation complexities and design cost. New trends and future directions including selective multirate filter banks, filter banks with nonuniform decimation ratio, filter banks with discrete coefficients, multidimensional filter banks, nonlinear filter banks and data dependent decompositions are presented. Finally, some selected applications of multirate filter banks are described; namely transmultiplexers, speech and image coding, analog voice privacy systems and wavelets based signal and image compression.