Faults are unavoidable for in-orbit satellites. They are able to be detected by carefully analyzing the telemetry data from these satellites. It is obvious that a good performance of anomaly detection algorithm is very important. This paper presents a new approach which takes the advantage of the similarity of preceding telemetry data to detect current anomaly. Because of its adaptive checking criteria, it can find some faults that cannot be found by some traditional methods. Most of important thing of this algorithm is that no prior knowledge about any particular satellite is needed except telemetry data themselves. In order to prove its availability, we use this new approach to process telemetry data from a China on-orbit satellite in 2012 and a fault is detected instantaneously.