This paper evaluates the performance of four clustering algorithms—Markov Clustering (MCL), Restricted Neighborhood Search Clustering (RNSC), Super Paramagnetic Clustering (SPC), and Molecular Complex Detection (MCODE)—for extracting protein complexes from protein-protein interaction networks. The evaluation is conducted using a test graph based on 220 annotated complexes from the MIPS database and altered graphs by adding or removing edges. The algorithms' sensitivity, positive predictive value (PPV), and geometric accuracy are assessed, and their robustness to false positives and false negatives is analyzed. MCL is found to be the most robust algorithm, with RNSC being more sensitive to edge deletion but less affected by suboptimal parameter settings. The algorithms are then applied to six high-throughput interaction networks from yeast, and their performance is compared with the annotated complexes. MCL consistently outperforms the other algorithms, particularly in terms of general performance and the contrast between real and permuted clusters. The study highlights the importance of parameter optimization and robustness analysis in clustering algorithms for extracting functional modules from protein interaction networks.This paper evaluates the performance of four clustering algorithms—Markov Clustering (MCL), Restricted Neighborhood Search Clustering (RNSC), Super Paramagnetic Clustering (SPC), and Molecular Complex Detection (MCODE)—for extracting protein complexes from protein-protein interaction networks. The evaluation is conducted using a test graph based on 220 annotated complexes from the MIPS database and altered graphs by adding or removing edges. The algorithms' sensitivity, positive predictive value (PPV), and geometric accuracy are assessed, and their robustness to false positives and false negatives is analyzed. MCL is found to be the most robust algorithm, with RNSC being more sensitive to edge deletion but less affected by suboptimal parameter settings. The algorithms are then applied to six high-throughput interaction networks from yeast, and their performance is compared with the annotated complexes. MCL consistently outperforms the other algorithms, particularly in terms of general performance and the contrast between real and permuted clusters. The study highlights the importance of parameter optimization and robustness analysis in clustering algorithms for extracting functional modules from protein interaction networks.