The pattern reduction (PR) algorithm we proposed previously, which works by eliminating patterns that are unlikely to change their membership during the convergence process, is obviously one of the most efficient methods for reducing the computation time of clustering algorithms. However, it is limited to problems with solutions that can be binary or integer encoded, such as combinatorial optimization problems. As such, this study is aimed at developing a new pattern reduction algorithm, called pattern reduction over continuous space, to get rid of this limitation. Like the PR, the proposed algorithm consists of two operators: detection and compression. Unlike the PR, the detection operator is divided into two steps. The first step is aimed at finding out subsolutions that can be considered as the candidate subsolutions for compression. The second step is performed to ensure that the candidate subsolutions have reached the final state so that any further computation is eventually a waste and thus can be compressed. To evaluate the performance of the proposed algorithm, we apply it to metaheuristics for clustering.