Aiming at the lack of accuracy, sensitivity and low time efficiency of protein function module mining methods based on spectral clustering and fuzzy C-means (FCM) clustering, a new algorithm named FSC-FM (functional module mining in uncertain PPI network based on fuzzy spectral clustering) was proposed. In the clustering process, in order to overcome the effect of false positives on the experimental results, the uncertain protein-protein interaction (PPI) network was constructed, in which each protein-protein interaction was assigned with an existence probability by using edge aggregation coefficient. At the same time, FEC (flow distance of edge clustering coefficient) measure was proposed to solve the problem that the spectral clustering is sensitive to the scale parameters in similarity matrix. Furthermore, based on density theory, a probability clustering center strategy was used to design an optimal selection method (DPCS) to improve accuracy and time efficiency of the algorithm. Finally, an improved EED (edge-expected density) metric was studied to filter out the functional modules to improve the precision of the algorithm. We compared our FSC-FM approach on yeast PPI data to the state-of-the-art functional modules prediction algorithms: CDUN, DCU, EA and MGPPA. The experimental results show the superiority of the FSC-FM algorithm in accuracy, sensitivity and time efficiency.