The efficiency of an algorithm depends greatly on the data structure adopted in practice, while the processing of useless data can cause not only much waste of memory but also much waste of time. Hence to eliminate redundant information is one of the main focuses in the research of algorithms, and the issue has received extensive attention in real-time field (especially those based on dense/continuous time semantics) for many years. After investigating existing problems in finite representation and operations, and by analyzing the dependence relation in information, a method is presented for eliminating redundant information in representation of dense time, which is developed based on an modification of the procedure of ‘Normalization’. The correctness of the method is proved and the efficiency is tested by experiences.