Abstract: Aiming at possible failure in software system, a novel software fault tolerance and fault rectification approach is proposed based on data flow abnomity supervision, software watch dog technique and rollback recover technique. This approach backups the object program regularly, establishes data flow analysis model by extracting a group of related variables in object program, and filtrates ‘outliers’ from the data flow by using data flow abnormity detection method. Implementation framework, operation procedure for above software fault tolerance and fault rectification strategy are presented. Binary regression model based on least square-support vector machine and ‘outliers’ detection algorithm are also studied. Simulation results demonstrate the correctness of the binary regression model and the validity of the ‘outliers’ detection algorithm.