In this paper, the checkpointing & rollback mechanism is studied deeply, and a new approach for SFT (software fault tolerance) is presented, which is based on the DFA (data flow analysis). The authors introduce the SFT technology at first, and point out that the peculiarity of data is the ultimate reason and also the final result of the software faults in control systems, so it is very necessary to adopt a powerful measure for data fault tolerance. Then, they discuss the applications of the DFA technique in the SFT in details, and give two theorems, one is for the minimal sufficient rollback point and another is for the checkpoint data set. Besides, they give the sufficient condition for the validation of the definition of a rollback block. They also use the telecommunication system as an example to illustrate the usage of this method. It is shown that only two checkpoints are needed to be maintained dynamically in their solution. The method presented in this paper can be used widely by extending simply.