Circular permutation connects the N and C termini of a protein and concurrently cleaves elsewhere in the chain, providing an important mechanism for generating novel protein fold and functions. However, their in genomes is unknown because current detection methods can miss many occurrences, mistaking random repeats as circular permutation. Here we develop a method for detecting circularly permuted proteins from structural comparison. Sequence order independent alignment of protein structures can be regarded as a special case of the maximum-weight independent set problem, which is known to be computationally hard. We develop an efficient approximation algorithm by repeatedly solving relaxations of an appropriate intermediate integer programming formulation, we show that the approximation ratio is much better than the theoretical worst case ratio of r=1/4. Circularly permuted proteins reported in literature can be identified rapidly with our method, while they escape the detection by publicly available servers for structural alignment.