One of the major challenges in molecular biology is to understand the precise mechanism by which gene expression is regulated. Reconstruction of transcription networks is essential to modeling this mechanism. In this paper, we describe a novel approach for building transcription networks from transcription modules by combining expression profile correlations with probabilistic element assessment. To demonstrate its performance, we systematically tested it on 27 transcription modules and reconstructed the transcription network for 6 transcription factors and 15 genes involved in the yeast cell cycle. The experimental results show that our combinatorial approach can better filter false positives to increase the selectivity in prediction of target genes. The regulatory control relationships described by the network reconstructed also mostly agree with those in earlier studies.