Identification of the interactions between molecular entities within cells is the key to understanding the biological processes involved. Unfortunately, it is difficult to identify these interactions entirely by experiments. Although numerous methods have been developed for inferring gene/protein regulatory networks from expression data, reliable network inference from gene/protein expression data remains an unsolved problem. Recently, a novel method, Divergence Weighted Independence Graphs (DWIG), was developed. A simulated data set with 160 virtual animals was generated from the mathematical model of insulin signaling pathway to evaluate the performance of DWIG. This simulated data characterized both the within-individual and between-individual variabilities, which other widely used public simulation data, such as the dream challenge, did not mimic fully. The performance of three reverse engineering methods, ARACNE (Algorithm for the Reconstruction of Accurate Cellular Networks), banjo (Bayesian network with java objects), and DWIG, were compared based on these simulated data. The Area Under Curve (AUC) of receiver operating characteristic (roc) curve showed that DWIG outperformed ARACNE and banjo. ARACNE uses the marginal mutual information, while DWIG uses the conditional mutual information, which could be the reason for DWIG’s superior performance over ARACNE. After prefiltering some weak links out by ARACNE, DWIG was applied to a protein dataset consisting of cytokines and chemokines that were measured in bronchoalveolar lavage fluid (BALF) of female a/j mice exposed to cigarette mainstream smoke for 3 and 5 months. An association network with 25 cytokines/chemokines was built.