The availability of complete genomic sequences and microarray expression data calls for computational methods for characterizing transcriptome, the complete collection of alternative transcription units (ATU) and complete transcription units (CTU). Though numerous computational methods have been developed for prediction of operons (CTU), none of existing computational methods can deal with ATU. We present a new computational method for ATU prediction. The ATU was predicted based on variance of fold changes of expression level and intergenic distance. Then, CTU and ATU were combined to form the transcriptome of Escherichia coli. The alternative TU predictor achieves 93% prediction accuracy in estimating presence of ATU. The percentage of known ATUs correctly predicted and known single-gene CTU correctly predicted are 84.3% and 80.43% respectively. About 91.94% of transcriptome (include CTU and ATU) from multiple-genes operons are correctly predicted.