This paper presents GENIE3, a new algorithm for inferring genetic regulatory networks (GRNs) from gene expression data using tree-based ensemble methods. The algorithm was the best performer in the DREAM4 In Silico Multifactorial challenge. GENIE3 decomposes the problem of inferring a regulatory network into p separate regression problems, where each problem predicts the expression pattern of one gene based on the expression patterns of all other genes. The importance of each gene in predicting the target gene's expression is used to infer potential regulatory links. These links are then aggregated to produce a ranking of interactions, which is used to reconstruct the network. GENIE3 does not assume any specific form of gene regulation, can handle combinatorial and non-linear interactions, and produces directed GRNs. It is fast and scalable, and performs well on both synthetic and real gene expression data. The algorithm uses feature selection with tree-based ensemble methods, making it adaptable to other types of genomic data and interactions. The paper also compares GENIE3 with existing algorithms on the E. coli regulatory network and shows that it performs well. The results show that GENIE3 can predict the direction of regulatory links, even with steady-state measurements, and outperforms other methods in some cases. The algorithm is available as open-source software.This paper presents GENIE3, a new algorithm for inferring genetic regulatory networks (GRNs) from gene expression data using tree-based ensemble methods. The algorithm was the best performer in the DREAM4 In Silico Multifactorial challenge. GENIE3 decomposes the problem of inferring a regulatory network into p separate regression problems, where each problem predicts the expression pattern of one gene based on the expression patterns of all other genes. The importance of each gene in predicting the target gene's expression is used to infer potential regulatory links. These links are then aggregated to produce a ranking of interactions, which is used to reconstruct the network. GENIE3 does not assume any specific form of gene regulation, can handle combinatorial and non-linear interactions, and produces directed GRNs. It is fast and scalable, and performs well on both synthetic and real gene expression data. The algorithm uses feature selection with tree-based ensemble methods, making it adaptable to other types of genomic data and interactions. The paper also compares GENIE3 with existing algorithms on the E. coli regulatory network and shows that it performs well. The results show that GENIE3 can predict the direction of regulatory links, even with steady-state measurements, and outperforms other methods in some cases. The algorithm is available as open-source software.