This paper presents high sample rate array architectures and sorting network-based architectures for computing both recursive and non-recursive median niters. The proposed array architectures have a sampling rate that is higher than that of existing architectures. This is achieved by pipelining the computations in each processor. While the non-recursive niters are pipelined by placing latches in the feed-forward paths, the recursive niters are restructured to create additional delays in the feedback paths, and then pipelined using the delays as latches. The proposed sorting network-based architectures are highly pipelined, and consist of fewer compare-swap units than existing architectures. The reduction in the number of compare-swap units is achieved by processing multiple outputs at the same time, and also by using Batcher's odd-even merge sort. The latency of these networks is reduced by building them with sorting units which sort 2 elements (sort-S) as well as 3 elements (sort-3) in 1 time unit.