Integration of various data sources (e.g., DNA sequence/motif information, mRNA expression data, and ChIP-chip data) has become one of the most important strategies for regulatory network inference. In this paper, a scheme called reliability analysis of network (RAN) is proposed for data integration via evaluating the reliability of regulatory nodes and edges. The RAN method addresses many challenges in data integration for network inference, in particular, the inconsistency between mRNA measurements and network connection information. The RAN approach is distinctly different from existing constraint factor analysis or multivariate regression based algorithms, which treat the network inference problem as a network parameter estimation problem. Rather, RAN provides the reliability measurement of regulatory nodes and edges in a statistical way for condition-dependent network inference. Computer simulation showed the effectiveness of RAN, both in regulatory node reliability evaluation and target identification. Furthermore, RAN has been applied to two microarray time-course datasets of breast cancer, and experimental results demonstrated that RAN can be used to identify not only the condition-dependent regulatory motifs but also their regulatory targets in transcriptional regulatory networks.