Commonly used in the areas of math and computer science, network propagation is a method that finds correlations within network data. Applied in the field of biology, network propagation shows powerful potential in making proximity predictions among genes, proteins, or other biological entities connected by a network structure. However, network propagation is sensitive toward perturbations in the data, making it unreliable when applied to incomplete or noisy biological network data. Therefore, performing confidence estimation on network propagation’s predictions is imperative for correctly interpreting the results. Currently, to our knowledge there is no existing method for confidence estimation in the context of network propagation. Simply applying a general-purpose confidence estimation method, such as permutation schemes, requires extensive computation. In this research we propose a novel approach that uses a recently described method, called the “knockoff filter,” to significantly reduce the computational cost of confidence estimation for network propagation. The knockoff filter is a method to perform false discovery rate (FDR) control for variable selection tasks. In a linear system that generates predictive system responses, the knockoff filter can be used to manufacture a set of knockoff variables given the original system variables. These knockoff variables can then serve as a negative control to help identify the truly important system responses. By applying the knockoff filter to network propagation-based biological proximity prediction algorithms, we are able to generate a knockoff network based on the original biological network. Then by comparing the network propagation results generated by both the original network and knockoff network, we are able to compute a confidence estimate for all the network propagation response variables. This approach provides fast and reliable confidence estimation.