Many computational methods have been developed to identify condition-specific transcription regulatory modules through sequence analysis and gene expression profiling. However, both gene expression data and motif binding data are noisy sources for regulatory module identification, which often results in many false positives in practice. In this paper, we propose a multi-level regulatory module identification method to discover significantly and stably enriched motif sets and their regulated gene modules. Specifically, motif binding strengths and gene expression profiles are integrated through support vector regression. Hypothesis testing is followed to discover significant regulatory modules. Finally, a multi-level procedure is designed to facilitate the identification of reliable regulatory modules. The experimental results on a breast cancer time course microarray data set show that the proposed method can successfully identify the significant and reliable regulatory modules at different conditions, which may provide important insights to the pathways related to breast cancer.